{ "transformer_blocks.0.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.01204797625541687, "diagnostic_nrmse": 0.01200854778289795, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.01646002009510994, "block-scale-search": 0.015654856339097023, "gptq": 0.01204797625541687 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.02176886983215809 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.018545852974057198 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.015424375422298908 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.021637562662363052 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.018992995843291283 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.01564367115497589 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.017608100548386574 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.016173778101801872 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.01271089818328619 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.01646002009510994 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.015654856339097023 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.01204797625541687 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.017391527071595192 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.016764748841524124 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.012740037404000759 } ] }, "transformer_blocks.0.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.013537188060581684, "diagnostic_nrmse": 0.01350016426295042, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.018655292689800262, "block-scale-search": 0.017673837020993233, "gptq": 0.013537188060581684 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.025481587275862694 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.021411621943116188 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.01775461807847023 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.02686687745153904 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.022570423781871796 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.018895704299211502 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.020305711776018143 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0182354673743248 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.014327256940305233 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.018655292689800262 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.017673837020993233 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.013537188060581684 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.020127693191170692 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.019346458837389946 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.014629285782575607 } ] }, "transformer_blocks.0.attn.to_v": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.030171392485499382, "diagnostic_nrmse": 0.03006247989833355, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04230092093348503, "block-scale-search": 0.04014577716588974, "gptq": 0.030171392485499382 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.05944820120930672 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.050177011638879776 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.041222307831048965 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05151767283678055 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0446167029440403 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03528943657875061 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04371869936585426 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.040114425122737885 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.030445391312241554 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04230092093348503 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04014577716588974 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.030171392485499382 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04486122727394104 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04316646605730057 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.032136525958776474 } ] }, "transformer_blocks.0.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.00714501878246665, "diagnostic_nrmse": 0.007025218568742275, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.009085649624466896, "block-scale-search": 0.0084769818931818, "gptq": 0.00714501878246665 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.009085649624466896 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0084769818931818 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.00714501878246665 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.009889854118227959 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.009119739755988121 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.007636731956154108 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.009516339749097824 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.008869816549122334 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0075789145193994045 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.009048228152096272 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.008553930558264256 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.007222309708595276 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.009027894586324692 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.008534833788871765 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.007219438906759024 } ] }, "transformer_blocks.0.img_mlp.proj": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04754091799259186, "diagnostic_nrmse": 0.04759332165122032, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06611739844083786, "block-scale-search": 0.062401413917541504, "gptq": 0.04754091799259186 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07857050746679306 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06992132216691971 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05641084909439087 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07130235433578491 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06476150453090668 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05060281977057457 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0669272392988205 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06231803074479103 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04772021248936653 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06611739844083786 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.062401413917541504 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04754091799259186 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06742769479751587 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0640890970826149 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04871291294693947 } ] }, "transformer_blocks.0.img_mlp.out": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.009401517920196056, "diagnostic_nrmse": 0.009935448877513409, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.011858324520289898, "block-scale-search": 0.011179910972714424, "gptq": 0.009401517920196056 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.01416892372071743 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.012253424152731895 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.010662768967449665 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.016328144818544388 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.014362838119268417 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.012434013187885284 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.012965780682861805 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.01187590416520834 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.010155302472412586 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.011858324520289898 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.011179910972714424 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.009401517920196056 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.012431219220161438 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.011910024099051952 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.009891637600958347 } ] }, "transformer_blocks.0.img_mlp.gate_layer": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.037483543157577515, "diagnostic_nrmse": 0.037637174129486084, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05210251733660698, "block-scale-search": 0.049135107547044754, "gptq": 0.037483543157577515 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0617147721350193 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.054840296506881714 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.044268619269132614 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.059090811759233475 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.053137023001909256 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04227127134799957 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05345984548330307 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.049534618854522705 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03820747882127762 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.05210251733660698 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.049135107547044754 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.037483543157577515 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.05318917706608772 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05055680871009827 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03848337382078171 } ] }, "transformer_blocks.1.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.01394839957356453, "diagnostic_nrmse": 0.013950744643807411, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.018765447661280632, "block-scale-search": 0.017707204446196556, "gptq": 0.01394839957356453 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.026817090809345245 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02223741076886654 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.01892644353210926 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.027167588472366333 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.022277729585766792 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.018876533955335617 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.020884601399302483 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.01866086572408676 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.015123274177312851 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.018765447661280632 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.017707204446196556 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.01394839957356453 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.01913086138665676 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.018293006345629692 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.014272979460656643 } ] }, "transformer_blocks.1.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.010926381684839725, "diagnostic_nrmse": 0.010980265215039253, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.014479911886155605, "block-scale-search": 0.013681148178875446, "gptq": 0.010926381684839725 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.021413585171103477 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.017297767102718353 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.014825601130723953 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.02187735214829445 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.01755620539188385 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.015024056658148766 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.016296638175845146 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.014572548680007458 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.011957862414419651 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.014479911886155605 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.013681148178875446 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.010926381684839725 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.01474265567958355 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.014088873751461506 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.011163771152496338 } ] }, "transformer_blocks.1.attn.to_v": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.02161742001771927, "diagnostic_nrmse": 0.021571993827819824, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.029608849436044693, "block-scale-search": 0.027858957648277283, "gptq": 0.02161742001771927 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04319457709789276 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03695789724588394 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03126239776611328 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03504988178610802 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.030904874205589294 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.025038115680217743 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03060719184577465 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028095383197069168 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.022006424143910408 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.029608849436044693 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.027858957648277283 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.02161742001771927 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.030450232326984406 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028964553028345108 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02241934835910797 } ] }, "transformer_blocks.1.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03111184388399124, "diagnostic_nrmse": 0.03133508563041687, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04202541708946228, "block-scale-search": 0.039323002099990845, "gptq": 0.03111184388399124 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04202541708946228 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.039323002099990845 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03111184388399124 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04277224466204643 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04006081819534302 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03175514191389084 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.042026087641716 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.039409954100847244 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03120107389986515 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04193828999996185 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.039402008056640625 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.031219685450196266 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.042318619787693024 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03984510898590088 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.031465813517570496 } ] }, "transformer_blocks.1.img_mlp.proj": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.043689798563718796, "diagnostic_nrmse": 0.043875470757484436, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05941176787018776, "block-scale-search": 0.05604563653469086, "gptq": 0.043689798563718796 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06619253009557724 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06036742776632309 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0488298162817955 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06256561726331711 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.057578179985284805 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.045585229992866516 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05990685895085335 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.055951911956071854 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04377031326293945 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.05941176787018776 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05604563653469086 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.043689798563718796 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06043759733438492 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.057370997965335846 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.044651735574007034 } ] }, "transformer_blocks.1.img_mlp.out": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.027740005403757095, "diagnostic_nrmse": 0.028688345104455948, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03577647730708122, "block-scale-search": 0.03379351645708084, "gptq": 0.027740005403757095 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04027485474944115 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03457673266530037 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.029407264664769173 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04546516016125679 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0392107293009758 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.033699966967105865 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03718152269721031 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03371473029255867 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.028116459026932716 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03577647730708122 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03379351645708084 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.027740005403757095 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04131810739636421 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.039717160165309906 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03211037442088127 } ] }, "transformer_blocks.1.img_mlp.gate_layer": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03621044009923935, "diagnostic_nrmse": 0.03669155761599541, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04924405738711357, "block-scale-search": 0.04640940576791763, "gptq": 0.03621044009923935 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.054762307554483414 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.049869976937770844 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04036976024508476 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.054076533764600754 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.049366921186447144 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.039680808782577515 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.050218649208545685 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04668508097529411 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.036707378923892975 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04924405738711357 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04640940576791763 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03621044009923935 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.050014667212963104 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.047461967915296555 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03698094189167023 } ] }, "transformer_blocks.2.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.021981237456202507, "diagnostic_nrmse": 0.02195565775036812, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.02921803668141365, "block-scale-search": 0.02755444310605526, "gptq": 0.021981237456202507 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03656823933124542 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03160259500145912 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.026668943464756012 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03489794209599495 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03073328360915184 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.02569129504263401 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03031611070036888 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.027914661914110184 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.022521087899804115 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.02921803668141365 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02755444310605526 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.021981237456202507 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03000529855489731 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028520645573735237 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.022701356559991837 } ] }, "transformer_blocks.2.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.018935570493340492, "diagnostic_nrmse": 0.018821358680725098, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.025102630257606506, "block-scale-search": 0.023600606247782707, "gptq": 0.018935570493340492 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03168255090713501 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02715022675693035 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.02297705039381981 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03268687054514885 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02781646139919758 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.023633649572730064 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.02675865963101387 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.024310849606990814 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.019821541383862495 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.025102630257606506 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.023600606247782707 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.018935570493340492 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.02555924467742443 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.024355849251151085 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.01944747194647789 } ] }, "transformer_blocks.2.attn.to_v": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03664879873394966, "diagnostic_nrmse": 0.03635368123650551, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04890240356326103, "block-scale-search": 0.04609983041882515, "gptq": 0.03664879873394966 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.061056625097990036 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05467776209115982 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04585718363523483 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05313146486878395 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.048465169966220856 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03916709125041962 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.049737587571144104 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04621801897883415 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03685003146529198 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04890240356326103 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04609983041882515 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03664879873394966 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.05011262744665146 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04756683111190796 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.037776898592710495 } ] }, "transformer_blocks.2.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05030475929379463, "diagnostic_nrmse": 0.0508437305688858, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06648268550634384, "block-scale-search": 0.0625414252281189, "gptq": 0.05030475929379463 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06710038334131241 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06273052096366882 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.050436291843652725 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06743969023227692 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06312650442123413 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05079549923539162 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06653054058551788 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06259015947580338 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.050371456891298294 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06648268550634384 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0625414252281189 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05030475929379463 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.067202128469944 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0632706955075264 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05083747208118439 } ] }, "transformer_blocks.2.img_mlp.proj": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04816805571317673, "diagnostic_nrmse": 0.048174601048231125, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06423712521791458, "block-scale-search": 0.06020296737551689, "gptq": 0.04816805571317673 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06792011857032776 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06281817704439163 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05119040235877037 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06629643589258194 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.061547428369522095 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.049702830612659454 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06423712521791458 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06020296737551689 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04816805571317673 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0639452114701271 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0604105181992054 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04822026565670967 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0649244487285614 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.061635758727788925 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.049085889011621475 } ] }, "transformer_blocks.2.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04337630793452263, "diagnostic_nrmse": 0.042701732367277145, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05610310286283493, "block-scale-search": 0.0504273995757103, "gptq": 0.04337630793452263 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.05610310286283493 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0504273995757103 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04337630793452263 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.060776688158512115 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05505768954753876 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04763484001159668 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05521643906831741 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.051255591213703156 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04359273985028267 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.05513974651694298 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05213101953268051 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04408177733421326 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06110258400440216 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05816500261425972 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.048886265605688095 } ] }, "transformer_blocks.2.img_mlp.gate_layer": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04238096997141838, "diagnostic_nrmse": 0.0424441322684288, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05625486001372337, "block-scale-search": 0.053075987845659256, "gptq": 0.04238096997141838 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.05962752923369408 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.055089179426431656 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04494895786046982 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0607169084250927 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0560695081949234 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04578891769051552 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05704871192574501 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05335201323032379 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.042834505438804626 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.05625486001372337 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.053075987845659256 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04238096997141838 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.05700397863984108 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.054055534303188324 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04311423748731613 } ] }, "transformer_blocks.3.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.022138092666864395, "diagnostic_nrmse": 0.021980702877044678, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.028738584369421005, "block-scale-search": 0.0270598866045475, "gptq": 0.022138092666864395 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03348654508590698 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02973177656531334 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.025240851566195488 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0346282497048378 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.030692771077156067 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.026188882067799568 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0299223642796278 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02751806005835533 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.022755661979317665 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.028738584369421005 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0270598866045475 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.022138092666864395 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0292659979313612 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02783740498125553 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02270531840622425 } ] }, "transformer_blocks.3.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.019831806421279907, "diagnostic_nrmse": 0.01971651054918766, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.025674965232610703, "block-scale-search": 0.02418961562216282, "gptq": 0.019831806421279907 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.030261749401688576 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.026619771495461464 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.022655807435512543 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0329136960208416 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02864447794854641 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.024628983810544014 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.027358919382095337 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.024957217276096344 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.020809412002563477 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.025674965232610703 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02418961562216282 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.019831806421279907 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.026080023497343063 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02479347586631775 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.020272476598620415 } ] }, "transformer_blocks.3.attn.to_v": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04759616032242775, "diagnostic_nrmse": 0.047322869300842285, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06196428835391998, "block-scale-search": 0.05839768424630165, "gptq": 0.04759616032242775 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0722498744726181 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0657794326543808 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05568607524037361 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06477593630552292 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05965502932667732 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0490090437233448 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06244232878088951 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05831347405910492 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.047602176666259766 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06196428835391998 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05839768424630165 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04759616032242775 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0633581355214119 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06019492819905281 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0490085668861866 } ] }, "transformer_blocks.3.attn.to_out.0": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06744234263896942, "diagnostic_nrmse": 0.06780756264925003, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0865737572312355, "block-scale-search": 0.08140603452920914, "gptq": 0.06744234263896942 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08828279376029968 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08284041285514832 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06871347874403 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08828619867563248 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0829373449087143 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06887225806713104 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0865737572312355 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08140603452920914 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06744234263896942 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08693907409906387 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08189257234334946 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06787656992673874 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08770983666181564 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08261966705322266 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06810154765844345 } ] }, "transformer_blocks.3.img_mlp.proj": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05244312807917595, "diagnostic_nrmse": 0.05254223570227623, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0679984912276268, "block-scale-search": 0.06390328705310822, "gptq": 0.05244312807917595 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06972932070493698 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0647692158818245 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05366722494363785 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06925159692764282 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06469663232564926 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05335753411054611 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0679984912276268 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06390328705310822 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05244312807917595 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06842752546072006 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06468137353658676 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.052892763167619705 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07030943781137466 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06669221073389053 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.054413143545389175 } ] }, "transformer_blocks.3.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05180466175079346, "diagnostic_nrmse": 0.0522477924823761, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06568221747875214, "block-scale-search": 0.06016981601715088, "gptq": 0.05180466175079346 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06568221747875214 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06016981601715088 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05180466175079346 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06810066848993301 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06314567476511002 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05449282377958298 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06556305289268494 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.061671093106269836 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.052928339689970016 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06815643608570099 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06454421579837799 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05524466186761856 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07621351629495621 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0725078359246254 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06144724786281586 } ] }, "transformer_blocks.3.img_mlp.gate_layer": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04803672060370445, "diagnostic_nrmse": 0.04827938601374626, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06226586923003197, "block-scale-search": 0.058513518422842026, "gptq": 0.04803672060370445 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06339474767446518 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.058842793107032776 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.048819709569215775 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06434424221515656 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05993283912539482 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.049632083624601364 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06226586923003197 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.058513518422842026 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04803672060370445 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06236245483160019 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05888254940509796 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04822039604187012 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06388819962739944 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.060586925595998764 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04944350942969322 } ] }, "transformer_blocks.4.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0300404354929924, "diagnostic_nrmse": 0.029558304697275162, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03823619708418846, "block-scale-search": 0.035944364964962006, "gptq": 0.0300404354929924 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04092825949192047 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03717109560966492 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03185500577092171 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.044716447591781616 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04025585576891899 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03488606587052345 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.039331238716840744 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03637935593724251 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.030811510980129242 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03823619708418846 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.035944364964962006 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0300404354929924 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03949086740612984 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03737816959619522 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.031074505299329758 } ] }, "transformer_blocks.4.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.02668287232518196, "diagnostic_nrmse": 0.02607416920363903, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.033966314047575, "block-scale-search": 0.0319070890545845, "gptq": 0.02668287232518196 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03670234978199005 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03302500769495964 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.028407752513885498 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03966754302382469 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03538845479488373 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.030690597370266914 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0350063182413578 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03221186250448227 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.027283960953354836 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.033966314047575 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0319070890545845 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.02668287232518196 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03497125953435898 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03315373882651329 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02756226435303688 } ] }, "transformer_blocks.4.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.055736590176820755, "diagnostic_nrmse": 0.055585023015737534, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07122404128313065, "block-scale-search": 0.06668663769960403, "gptq": 0.055736590176820755 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07863883674144745 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0723549872636795 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06200990825891495 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07229160517454147 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0670405924320221 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.056286342442035675 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07122404128313065 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06668663769960403 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.055736590176820755 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0725838914513588 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0685015469789505 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0571107417345047 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07624177634716034 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07226123660802841 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05989644676446915 } ] }, "transformer_blocks.4.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06598164141178131, "diagnostic_nrmse": 0.06529822945594788, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08462206274271011, "block-scale-search": 0.07946640253067017, "gptq": 0.06598164141178131 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08462206274271011 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07946640253067017 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06598164141178131 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08618468791246414 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08095619082450867 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0673082247376442 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0857187956571579 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0804314836859703 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06696391850709915 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08612225949764252 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08111876249313354 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06750041246414185 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08737868070602417 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08200975507497787 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06815282255411148 } ] }, "transformer_blocks.4.img_mlp.proj": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05826263129711151, "diagnostic_nrmse": 0.058187585324048996, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07371167838573456, "block-scale-search": 0.06938529014587402, "gptq": 0.05826263129711151 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07479724287986755 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06970960646867752 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0589606799185276 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07463854551315308 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06973093003034592 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05880667269229889 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07371167838573456 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06938529014587402 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05826263129711151 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07435997575521469 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07019958645105362 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.058856263756752014 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07607226073741913 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07200721651315689 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06015254929661751 } ] }, "transformer_blocks.4.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06170419603586197, "diagnostic_nrmse": 0.06021061912178993, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07635167241096497, "block-scale-search": 0.07047370076179504, "gptq": 0.06170419603586197 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07635167241096497 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07047370076179504 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06170419603586197 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07854028046131134 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07304005324840546 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06412827223539352 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07653442025184631 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07213505357503891 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0629771426320076 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07952794432640076 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07514297217130661 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06559034436941147 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08759711682796478 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08312726020812988 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07217446714639664 } ] }, "transformer_blocks.4.img_mlp.gate_layer": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05295782908797264, "diagnostic_nrmse": 0.052818503230810165, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0670795664191246, "block-scale-search": 0.06300381571054459, "gptq": 0.05295782908797264 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06735213845968246 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06272587925195694 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05307267978787422 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06878870725631714 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06419064104557037 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05420884117484093 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0670795664191246 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06300381571054459 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05295782908797264 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06705165654420853 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06331271678209305 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.053061023354530334 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06837578862905502 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0647755116224289 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05411967635154724 } ] }, "transformer_blocks.5.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03141239285469055, "diagnostic_nrmse": 0.030794253572821617, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.039393533021211624, "block-scale-search": 0.03700704500079155, "gptq": 0.03141239285469055 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.042088016867637634 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.038446951657533646 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03330627456307411 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04434455186128616 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04025657847523689 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03510138392448425 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04032142460346222 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03731367364525795 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03195998817682266 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.039393533021211624 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03700704500079155 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03141239285469055 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.040316142141819 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03819650784134865 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03231631964445114 } ] }, "transformer_blocks.5.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.02782702073454857, "diagnostic_nrmse": 0.027161594480276108, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03490222990512848, "block-scale-search": 0.03280038759112358, "gptq": 0.02782702073454857 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03726322948932648 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.033935610204935074 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.029425565153360367 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04147852957248688 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0372912771999836 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.032744791358709335 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03621111065149307 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0334676131606102 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.028722155839204788 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03490222990512848 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03280038759112358 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.02782702073454857 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.035513028502464294 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03363391384482384 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.028488531708717346 } ] }, "transformer_blocks.5.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06094220653176308, "diagnostic_nrmse": 0.06027190387248993, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07655340433120728, "block-scale-search": 0.07166853547096252, "gptq": 0.06094220653176308 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08270388841629028 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07633974403142929 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0659734234213829 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07824035733938217 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0724458172917366 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06186099722981453 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07655340433120728 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07166853547096252 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06094220653176308 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07722868025302887 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.072716623544693 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06162859871983528 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08001024276018143 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07571662962436676 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06392787396907806 } ] }, "transformer_blocks.5.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07536669075489044, "diagnostic_nrmse": 0.074112169444561, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09603533893823624, "block-scale-search": 0.09004221856594086, "gptq": 0.07536669075489044 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09603533893823624 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09004221856594086 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07536669075489044 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09830553829669952 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09229818731546402 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07751637697219849 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09722494333982468 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09135669469833374 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07656417042016983 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0976642593741417 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09170128405094147 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0769699215888977 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0987829715013504 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09257998317480087 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07746776193380356 } ] }, "transformer_blocks.5.img_mlp.proj": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05958409607410431, "diagnostic_nrmse": 0.05924244225025177, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.074283666908741, "block-scale-search": 0.06981762498617172, "gptq": 0.05958409607410431 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07483546435832977 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06980547308921814 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05988704413175583 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07503996789455414 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07016336917877197 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06006268411874771 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.074283666908741 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06981762498617172 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05958409607410431 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07476256787776947 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07059551775455475 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.060135368257761 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07633859664201736 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07228117436170578 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06139559671282768 } ] }, "transformer_blocks.5.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0698762983083725, "diagnostic_nrmse": 0.06754415482282639, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08645671606063843, "block-scale-search": 0.07959997653961182, "gptq": 0.0698762983083725 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08645671606063843 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07959997653961182 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0698762983083725 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08933093398809433 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0831090584397316 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07359062135219574 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08687667548656464 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0816100612282753 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07196871936321259 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09001241624355316 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08493799716234207 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07483195513486862 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09912475198507309 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0936877503991127 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.08226865530014038 } ] }, "transformer_blocks.5.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05468739569187164, "diagnostic_nrmse": 0.05436067655682564, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06833047419786453, "block-scale-search": 0.06370331346988678, "gptq": 0.05468739569187164 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06833047419786453 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06370331346988678 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05468739569187164 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07000645250082016 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0653143972158432 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05591702461242676 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06839211285114288 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06428129225969315 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.054872091859579086 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0684383288025856 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06458493322134018 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.055009875446558 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06963726133108139 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06595009565353394 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.055995259433984756 } ] }, "transformer_blocks.6.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03288814052939415, "diagnostic_nrmse": 0.0323331356048584, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.040873024612665176, "block-scale-search": 0.03840687870979309, "gptq": 0.03288814052939415 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04316386580467224 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0396297350525856 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03450335934758186 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04562198743224144 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.041751932352781296 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03658193722367287 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04171079769730568 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0387602224946022 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.033376384526491165 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.040873024612665176 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03840687870979309 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03288814052939415 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04171507805585861 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.039472900331020355 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.033658966422080994 } ] }, "transformer_blocks.6.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03111395053565502, "diagnostic_nrmse": 0.03039770945906639, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03873226419091225, "block-scale-search": 0.03635359928011894, "gptq": 0.03111395053565502 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04107769578695297 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03753024712204933 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03270195424556732 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04365541785955429 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03978303447365761 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03493579477071762 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.039649445563554764 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03678194060921669 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03172332048416138 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03873226419091225 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03635359928011894 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03111395053565502 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.039333194494247437 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03724272549152374 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03174476698040962 } ] }, "transformer_blocks.6.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06554220616817474, "diagnostic_nrmse": 0.0651458203792572, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08161679655313492, "block-scale-search": 0.07663996517658234, "gptq": 0.06554220616817474 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08682935684919357 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08047489076852798 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07000675797462463 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0829399898648262 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07713094353675842 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0662781372666359 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08161679655313492 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07663996517658234 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06554220616817474 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08227669447660446 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07767964154481888 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06624375283718109 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08487895131111145 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08033818006515503 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06835969537496567 } ] }, "transformer_blocks.6.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08209089189767838, "diagnostic_nrmse": 0.08089428395032883, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.10471979528665543, "block-scale-search": 0.09806748479604721, "gptq": 0.08209089189767838 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10471979528665543 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09806748479604721 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08209089189767838 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.10617150366306305 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0995599627494812 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.08318842202425003 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.10510478913784027 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0988556519150734 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.08269987255334854 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.10605460405349731 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09953649342060089 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.08330971002578735 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10700326412916183 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.10059552639722824 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0839093029499054 } ] }, "transformer_blocks.6.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05947941914200783, "diagnostic_nrmse": 0.058947689831256866, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07338657230138779, "block-scale-search": 0.06853201240301132, "gptq": 0.05947941914200783 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07338657230138779 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06853201240301132 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05947941914200783 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07423052191734314 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06951287388801575 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06025853380560875 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07330673187971115 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06895826011896133 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05957278236746788 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.073734350502491 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06951171159744263 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05990365520119667 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07521138340234756 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07105880975723267 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06110260263085365 } ] }, "transformer_blocks.6.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07063374668359756, "diagnostic_nrmse": 0.0704704225063324, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08702941238880157, "block-scale-search": 0.08033306896686554, "gptq": 0.07063374668359756 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08702941238880157 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08033306896686554 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07063374668359756 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09144146740436554 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08510527014732361 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07577377557754517 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08822786808013916 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08276375383138657 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07316041737794876 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09077247977256775 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08554136753082275 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07534326612949371 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09925463050603867 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09356645494699478 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.08221606910228729 } ] }, "transformer_blocks.6.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05478712543845177, "diagnostic_nrmse": 0.05440879240632057, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06768565624952316, "block-scale-search": 0.06316740810871124, "gptq": 0.05478712543845177 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06768565624952316 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06316740810871124 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05478712543845177 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06909049302339554 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06450223177671432 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05590072646737099 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0677718073129654 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06360259652137756 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05497262626886368 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0679904893040657 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06399543583393097 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.055200546979904175 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06921134889125824 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06537635624408722 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.056216005235910416 } ] }, "transformer_blocks.7.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.034951597452163696, "diagnostic_nrmse": 0.03410928696393967, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04323725029826164, "block-scale-search": 0.04056887328624725, "gptq": 0.034951597452163696 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0453609861433506 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04175999388098717 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03651995584368706 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0477634035050869 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.043876517564058304 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.038589682430028915 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04403207078576088 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.041011445224285126 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03553716838359833 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04323725029826164 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04056887328624725 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.034951597452163696 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04389992356300354 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04147646203637123 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0356881283223629 } ] }, "transformer_blocks.7.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.033480264246463776, "diagnostic_nrmse": 0.032536134123802185, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04134387522935867, "block-scale-search": 0.03883988782763481, "gptq": 0.033480264246463776 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.043515514582395554 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03988891839981079 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03493328392505646 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04640170559287071 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04245446249842644 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.037425074726343155 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04235643893480301 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03933491185307503 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.034132037311792374 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04134387522935867 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03883988782763481 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.033480264246463776 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.041825998574495316 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03951752558350563 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03401000797748566 } ] }, "transformer_blocks.7.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06734803318977356, "diagnostic_nrmse": 0.06662296503782272, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08333755284547806, "block-scale-search": 0.07785684615373611, "gptq": 0.06734803318977356 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08851324021816254 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08201301842927933 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07159788906574249 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0847451388835907 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07854971289634705 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06816576421260834 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08333755284547806 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07785684615373611 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06734803318977356 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08380576223134995 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07893721014261246 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06812024861574173 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08654782176017761 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08193758130073547 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0704912319779396 } ] }, "transformer_blocks.7.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.09091086685657501, "diagnostic_nrmse": 0.09075409173965454, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.11696715652942657, "block-scale-search": 0.10959166288375854, "gptq": 0.09091086685657501 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.11696715652942657 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.10959166288375854 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.09091086685657501 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.1194533184170723 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11191193759441376 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.09300696849822998 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.11725836992263794 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11001721769571304 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.09132686257362366 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.11735693365335464 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11011373996734619 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.09167122095823288 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.1182856559753418 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11069634556770325 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.09217613190412521 } ] }, "transformer_blocks.7.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05875392258167267, "diagnostic_nrmse": 0.05795391649007797, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07170682400465012, "block-scale-search": 0.06702882796525955, "gptq": 0.05875392258167267 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07170682400465012 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06702882796525955 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05875392258167267 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07222610712051392 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06763533502817154 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05923997983336449 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0716647207736969 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06733034551143646 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05878368765115738 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07218609005212784 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06798582524061203 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05924437567591667 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07357831299304962 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06943365931510925 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.060367830097675323 } ] }, "transformer_blocks.7.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06075708195567131, "diagnostic_nrmse": 0.05955856665968895, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07464698702096939, "block-scale-search": 0.06913892179727554, "gptq": 0.06075708195567131 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07464698702096939 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06913892179727554 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06075708195567131 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08040221780538559 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0749465823173523 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06617937982082367 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0772828608751297 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07255218178033829 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06363657861948013 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07913050800561905 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0743958055973053 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06514639407396317 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0858645811676979 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08089792728424072 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07052711397409439 } ] }, "transformer_blocks.7.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.054100822657346725, "diagnostic_nrmse": 0.05346961319446564, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06610828638076782, "block-scale-search": 0.061784692108631134, "gptq": 0.054100822657346725 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06610828638076782 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.061784692108631134 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.054100822657346725 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06763416528701782 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06310409307479858 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05527864769101143 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.066353939473629 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.062243591994047165 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05431428551673889 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06648903340101242 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06258276849985123 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0545000396668911 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06768074631690979 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06385869532823563 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.055496618151664734 } ] }, "transformer_blocks.8.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03318856284022331, "diagnostic_nrmse": 0.03256601467728615, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04067879915237427, "block-scale-search": 0.038101740181446075, "gptq": 0.03318856284022331 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.042560890316963196 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.039243344217538834 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.034583304077386856 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.047159865498542786 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.043017130345106125 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.038290973752737045 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04195678234100342 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.038868408650159836 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03412432223558426 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04067879915237427 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.038101740181446075 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03318856284022331 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.041078995913267136 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03876323252916336 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03368967026472092 } ] }, "transformer_blocks.8.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03260636702179909, "diagnostic_nrmse": 0.03172523155808449, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.040004462003707886, "block-scale-search": 0.03741814196109772, "gptq": 0.03260636702179909 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.042044468224048615 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.038646847009658813 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.034044042229652405 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04561201110482216 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04165761545300484 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0370914451777935 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.041279472410678864 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03818893060088158 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.033496253192424774 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.040004462003707886 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03741814196109772 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03260636702179909 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.040220215916633606 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03793413192033768 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03298841789364815 } ] }, "transformer_blocks.8.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0706174299120903, "diagnostic_nrmse": 0.06976960599422455, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08647215366363525, "block-scale-search": 0.08108068257570267, "gptq": 0.0706174299120903 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09147592633962631 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08509717136621475 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07477977871894836 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08721673488616943 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08119792491197586 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07088298350572586 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08647215366363525 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08108068257570267 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0706174299120903 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08709094673395157 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08207550644874573 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07138817757368088 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08914213627576828 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08415870368480682 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07301408052444458 } ] }, "transformer_blocks.8.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.09470555931329727, "diagnostic_nrmse": 0.09487172961235046, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.1218191608786583, "block-scale-search": 0.1141551285982132, "gptq": 0.09470555931329727 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.1218191608786583 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1141551285982132 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.09470555931329727 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.12320323288440704 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11558172851800919 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.09606537967920303 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.12144888937473297 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11410610377788544 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0947384312748909 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.12182512879371643 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1142674908041954 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.09502647817134857 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.12337243556976318 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11555518954992294 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.09586285054683685 } ] }, "transformer_blocks.8.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.059872061014175415, "diagnostic_nrmse": 0.05900264158844948, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07268749922513962, "block-scale-search": 0.06798353046178818, "gptq": 0.059872061014175415 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07268749922513962 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06798353046178818 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.059872061014175415 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07337597012519836 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06865998357534409 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.060475870966911316 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07271324843168259 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06825145334005356 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05996515229344368 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07327645272016525 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06899397075176239 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06050686538219452 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0745888352394104 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07036034017801285 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.061595577746629715 } ] }, "transformer_blocks.8.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.056356921792030334, "diagnostic_nrmse": 0.054451778531074524, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0697629302740097, "block-scale-search": 0.06421709805727005, "gptq": 0.056356921792030334 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0697629302740097 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06421709805727005 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.056356921792030334 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07463224232196808 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06937366724014282 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.061221398413181305 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07181013375520706 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06723994016647339 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0589892640709877 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07366389036178589 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06916403025388718 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06045687571167946 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08013399690389633 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07543329894542694 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06559234857559204 } ] }, "transformer_blocks.8.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05422260984778404, "diagnostic_nrmse": 0.053331661969423294, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06585337966680527, "block-scale-search": 0.061611276119947433, "gptq": 0.05422260984778404 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06585337966680527 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.061611276119947433 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05422260984778404 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06729570776224136 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06278186291456223 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05530085787177086 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06619840115308762 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0620541125535965 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05450943857431412 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06631913781166077 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.062349844723939896 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05470149219036102 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06749803572893143 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06359734386205673 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.055633071810007095 } ] }, "transformer_blocks.9.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03728332370519638, "diagnostic_nrmse": 0.036881737411022186, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.045497722923755646, "block-scale-search": 0.0426328182220459, "gptq": 0.03728332370519638 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04718220606446266 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.043721843510866165 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.038611311465501785 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.050358790904283524 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.046513453125953674 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04132785275578499 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.046455223113298416 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04327433556318283 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.038046594709157944 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.045497722923755646 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0426328182220459 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03728332370519638 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04612591117620468 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04350410774350166 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03793732449412346 } ] }, "transformer_blocks.9.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03824923560023308, "diagnostic_nrmse": 0.03722543269395828, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0467325858771801, "block-scale-search": 0.04384714737534523, "gptq": 0.03824923560023308 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04874926805496216 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04495248571038246 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03970588371157646 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05071519687771797 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04683058708906174 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04152144864201546 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.047545112669467926 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04420280456542969 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.038786981254816055 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0467325858771801 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04384714737534523 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03824923560023308 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04717114940285683 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04447353258728981 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03876921162009239 } ] }, "transformer_blocks.9.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.071395643055439, "diagnostic_nrmse": 0.07051113992929459, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08702494949102402, "block-scale-search": 0.08147365599870682, "gptq": 0.071395643055439 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09120361506938934 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08498736470937729 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07495590299367905 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08802232891321182 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08176363259553909 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0718805268406868 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08702494949102402 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08147365599870682 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.071395643055439 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08761667460203171 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0825611799955368 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07224541157484055 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09026505053043365 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08522332459688187 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0742943063378334 } ] }, "transformer_blocks.9.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0965825542807579, "diagnostic_nrmse": 0.097048819065094, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.12521085143089294, "block-scale-search": 0.11746110767126083, "gptq": 0.0965825542807579 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.12521085143089294 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11746110767126083 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0965825542807579 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.12672339379787445 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1187390461564064 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0979095995426178 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.12526147067546844 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11729113757610321 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0968344435095787 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.12574324011802673 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11811220645904541 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.09768740832805634 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.1271447390317917 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11933085322380066 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.09836564213037491 } ] }, "transformer_blocks.9.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06083711236715317, "diagnostic_nrmse": 0.05960188806056976, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07350830733776093, "block-scale-search": 0.06884079426527023, "gptq": 0.06083711236715317 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07350830733776093 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06884079426527023 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06083711236715317 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07415865361690521 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06941530853509903 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06135431304574013 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07395051419734955 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06939798593521118 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0611741878092289 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07501285523176193 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07054194062948227 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06207556277513504 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07702157646417618 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07262644916772842 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06357605755329132 } ] }, "transformer_blocks.9.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06274417787790298, "diagnostic_nrmse": 0.061307404190301895, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07850395888090134, "block-scale-search": 0.07247290015220642, "gptq": 0.06274417787790298 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07850395888090134 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07247290015220642 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06274417787790298 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08383187651634216 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07844973355531693 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06821019947528839 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08062638342380524 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07599923014640808 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06575210392475128 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08368416130542755 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07923129200935364 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0683726891875267 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09326062351465225 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08816927671432495 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07537174224853516 } ] }, "transformer_blocks.9.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.055773116648197174, "diagnostic_nrmse": 0.054580576717853546, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06747160851955414, "block-scale-search": 0.06317103654146194, "gptq": 0.055773116648197174 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06747160851955414 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06317103654146194 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.055773116648197174 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06916963309049606 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06452580541372299 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05704237148165703 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06824959069490433 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06394446641206741 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05639313533902168 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06881820410490036 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06462252140045166 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05684063956141472 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07057436555624008 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06645940244197845 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05825505033135414 } ] }, "transformer_blocks.10.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.037456076592206955, "diagnostic_nrmse": 0.036694444715976715, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04559323564171791, "block-scale-search": 0.042741477489471436, "gptq": 0.037456076592206955 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04633346199989319 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04296621307730675 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03802802786231041 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04936906695365906 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04551747813820839 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0405539907515049 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04604316130280495 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.042860548943281174 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03779575601220131 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04559323564171791 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.042741477489471436 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.037456076592206955 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.047021009027957916 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.044301826506853104 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03862331062555313 } ] }, "transformer_blocks.10.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.037324290722608566, "diagnostic_nrmse": 0.036280203610658646, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.045569028705358505, "block-scale-search": 0.04268184304237366, "gptq": 0.037324290722608566 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.046463534235954285 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04287669062614441 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.038004741072654724 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04871173948049545 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0448940135538578 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.039972323924303055 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04580327495932579 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.042623985558748245 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03748354688286781 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.045569028705358505 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04268184304237366 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.037324290722608566 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.046633150428533554 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0438719168305397 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03827669844031334 } ] }, "transformer_blocks.10.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07489950209856033, "diagnostic_nrmse": 0.07374904304742813, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09111925959587097, "block-scale-search": 0.08527754247188568, "gptq": 0.07489950209856033 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09451527148485184 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08822253346443176 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07805082201957703 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09123179316520691 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08492755144834518 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07495376467704773 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09111925959587097 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08527754247188568 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07489950209856033 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09302549809217453 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08737539499998093 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07672110199928284 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09713701158761978 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0915812999010086 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07993960380554199 } ] }, "transformer_blocks.10.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.10148494690656662, "diagnostic_nrmse": 0.1023259237408638, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.13250578939914703, "block-scale-search": 0.12400633096694946, "gptq": 0.10148494690656662 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.13250578939914703 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12400633096694946 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.10148494690656662 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.1332578957080841 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1248689815402031 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.10228173434734344 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.13225340843200684 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1242215633392334 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.10190766304731369 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.13316620886325836 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12497636675834656 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.10240998864173889 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.1349971443414688 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12647564709186554 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.10347003489732742 } ] }, "transformer_blocks.10.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.061430059373378754, "diagnostic_nrmse": 0.060185886919498444, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.074025958776474, "block-scale-search": 0.0693536251783371, "gptq": 0.061430059373378754 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.074025958776474 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0693536251783371 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.061430059373378754 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07482913881540298 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07013004273176193 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.062074411660432816 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07486768811941147 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07026655226945877 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06209617108106613 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07619289308786392 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07171209156513214 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.063197560608387 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07855250686407089 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07401467114686966 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06494417041540146 } ] }, "transformer_blocks.10.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06744197756052017, "diagnostic_nrmse": 0.06539103388786316, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08494407683610916, "block-scale-search": 0.07847711443901062, "gptq": 0.06744197756052017 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08494407683610916 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07847711443901062 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06744197756052017 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08957405388355255 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08367644250392914 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07246793061494827 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08692701905965805 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08200927078723907 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0705334022641182 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09037750214338303 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08543986082077026 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07336574047803879 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09981927275657654 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09482364356517792 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.08059521019458771 } ] }, "transformer_blocks.10.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05763446167111397, "diagnostic_nrmse": 0.05636850744485855, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06949494779109955, "block-scale-search": 0.06510767340660095, "gptq": 0.05763446167111397 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06949494779109955 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06510767340660095 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05763446167111397 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07121918350458145 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06648053973913193 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05892516300082207 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07046401500701904 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06606529653072357 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05835307389497757 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07142037898302078 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06719156354665756 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05919326841831207 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07361311465501785 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06936389207839966 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0607975497841835 } ] }, "transformer_blocks.11.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03355240821838379, "diagnostic_nrmse": 0.033099737018346786, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04073972627520561, "block-scale-search": 0.03820149600505829, "gptq": 0.03355240821838379 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04174115136265755 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03865089267492294 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03427278995513916 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04546917974948883 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.041773803532123566 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03735150769352913 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.041639309376478195 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.038708075881004333 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03424068167805672 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04073972627520561 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03820149600505829 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03355240821838379 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.041799336671829224 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03936319798231125 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.034475091844797134 } ] }, "transformer_blocks.11.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03789547458291054, "diagnostic_nrmse": 0.03692049905657768, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04607049375772476, "block-scale-search": 0.04317709803581238, "gptq": 0.03789547458291054 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04738284647464752 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04373297467827797 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0387452058494091 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.049145158380270004 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04538371041417122 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0403977707028389 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.046487290412187576 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04335014522075653 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.038200750946998596 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04607049375772476 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04317709803581238 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03789547458291054 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04695778340101242 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04426444321870804 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03872481361031532 } ] }, "transformer_blocks.11.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0764651969075203, "diagnostic_nrmse": 0.07594986259937286, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09265657514333725, "block-scale-search": 0.08683153241872787, "gptq": 0.0764651969075203 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09609189629554749 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08969713002443314 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07945447415113449 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09301687031984329 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08674785494804382 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07651519030332565 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09265657514333725 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08683153241872787 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0764651969075203 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09410924464464188 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08865045756101608 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07783763855695724 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09764569997787476 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09223974496126175 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.08067181706428528 } ] }, "transformer_blocks.11.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.10521842539310455, "diagnostic_nrmse": 0.10661596059799194, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.13635976612567902, "block-scale-search": 0.12784059345722198, "gptq": 0.10521842539310455 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.13765911757946014 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1289641410112381 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.10570574551820755 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.13873916864395142 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12998263537883759 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.10687629878520966 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.13667668402194977 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12817516922950745 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.10532370954751968 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.13635976612567902 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12784059345722198 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.10521842539310455 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.13685841858386993 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12832742929458618 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.10528375953435898 } ] }, "transformer_blocks.11.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06189156696200371, "diagnostic_nrmse": 0.061123814433813095, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07427813857793808, "block-scale-search": 0.06964311748743057, "gptq": 0.06189156696200371 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07427813857793808 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06964311748743057 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06189156696200371 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07506309449672699 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07033678144216537 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06247629225254059 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07491113245487213 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07034534215927124 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06236230954527855 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07584988325834274 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07141391187906265 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06315334141254425 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07802104204893112 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0736079066991806 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06487880647182465 } ] }, "transformer_blocks.11.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06591105461120605, "diagnostic_nrmse": 0.06493569910526276, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08232971280813217, "block-scale-search": 0.07608294486999512, "gptq": 0.06591105461120605 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08232971280813217 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07608294486999512 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06591105461120605 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08703596144914627 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08099129796028137 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07071419060230255 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08464109897613525 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07909125089645386 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06853323429822922 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08666699379682541 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08137762546539307 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07048554718494415 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09324007481336594 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08806656301021576 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07594914734363556 } ] }, "transformer_blocks.11.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05782931670546532, "diagnostic_nrmse": 0.05712378770112991, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06952250748872757, "block-scale-search": 0.06510807573795319, "gptq": 0.05782931670546532 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06952250748872757 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06510807573795319 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05782931670546532 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07110273838043213 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06646494567394257 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.059064339846372604 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07041970640420914 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06606095284223557 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05855606496334076 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07099254429340363 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06677761673927307 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05908503383398056 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07282306998968124 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06866386532783508 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0604880191385746 } ] }, "transformer_blocks.12.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03785655274987221, "diagnostic_nrmse": 0.03743819519877434, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04565282166004181, "block-scale-search": 0.04282018542289734, "gptq": 0.03785655274987221 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04677373543381691 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.043414127081632614 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03869796171784401 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05096805468201637 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.047152433544397354 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04235853999853134 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04665354639291763 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04353845492005348 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0386715866625309 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04565282166004181 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04282018542289734 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03785655274987221 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04628957435488701 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04362758994102478 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03847559913992882 } ] }, "transformer_blocks.12.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03987666219472885, "diagnostic_nrmse": 0.03895050287246704, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.048124272376298904, "block-scale-search": 0.045103661715984344, "gptq": 0.03987666219472885 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04959384351968765 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04590987041592598 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04091480001807213 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.051955364644527435 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04798571765422821 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04298614338040352 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.048840735107660294 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04543593153357506 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04035599157214165 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.048124272376298904 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.045103661715984344 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03987666219472885 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.048826128244400024 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04597226157784462 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04055866226553917 } ] }, "transformer_blocks.12.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07483004778623581, "diagnostic_nrmse": 0.07379419356584549, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08985097706317902, "block-scale-search": 0.08436480164527893, "gptq": 0.07483004778623581 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09316520392894745 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08719168603420258 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07760827988386154 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09077794849872589 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08465155959129333 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07528334110975266 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08985097706317902 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08436480164527893 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07483004778623581 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0909079983830452 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08552763611078262 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07570730894804001 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09397263079881668 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08870232105255127 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0782325342297554 } ] }, "transformer_blocks.12.attn.to_out.0": { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.09953557699918747, "diagnostic_nrmse": 0.10188543796539307, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.12970443069934845, "block-scale-search": 0.12158212810754776, "gptq": 0.09953557699918747 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.13131476938724518 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12296442687511444 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.10081416368484497 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.1320856660604477 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12370702624320984 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.10148678719997406 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.13030026853084564 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12195362895727158 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.10013636946678162 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.12972810864448547 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12166977673768997 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.09970886260271072 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.12970443069934845 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12158212810754776 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.09953557699918747 } ] }, "transformer_blocks.12.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.060857683420181274, "diagnostic_nrmse": 0.060051847249269485, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07259713858366013, "block-scale-search": 0.06811422109603882, "gptq": 0.060857683420181274 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07259713858366013 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06811422109603882 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.060857683420181274 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07348519563674927 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06901576370000839 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06163134425878525 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07321184128522873 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0688512772321701 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.061376579105854034 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07398834079504013 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06973636895418167 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.062056414783000946 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07569117099046707 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07133221626281738 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06327628344297409 } ] }, "transformer_blocks.12.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06155862659215927, "diagnostic_nrmse": 0.06034066900610924, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07652650773525238, "block-scale-search": 0.07036387920379639, "gptq": 0.06155862659215927 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07652650773525238 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07036387920379639 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06155862659215927 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08246565610170364 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0761830285191536 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06724863499403 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07892277091741562 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0733029693365097 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0642724335193634 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07941094040870667 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07422328740358353 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06495123356580734 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08418460935354233 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07894902676343918 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06894926726818085 } ] }, "transformer_blocks.12.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.059681203216314316, "diagnostic_nrmse": 0.05891493335366249, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07122746855020523, "block-scale-search": 0.06677602231502533, "gptq": 0.059681203216314316 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07122746855020523 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06677602231502533 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.059681203216314316 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07258003205060959 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06795249879360199 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06071121618151665 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07189971953630447 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06761480122804642 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06024588644504547 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07257979363203049 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06833413988351822 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06078431382775307 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0740903988480568 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06983963400125504 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06192372739315033 } ] }, "transformer_blocks.13.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.032707832753658295, "diagnostic_nrmse": 0.032450735569000244, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03948400169610977, "block-scale-search": 0.03701634705066681, "gptq": 0.032707832753658295 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.040916163474321365 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03790086507797241 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.033743247389793396 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.043515559285879135 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04008040204644203 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.035909503698349 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04023859277367592 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0374499186873436 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03327213227748871 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03948400169610977 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03701634705066681 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.032707832753658295 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04008639603853226 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03770070895552635 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03325069695711136 } ] }, "transformer_blocks.13.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03402695432305336, "diagnostic_nrmse": 0.03327338397502899, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04108581319451332, "block-scale-search": 0.03857039660215378, "gptq": 0.03402695432305336 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04276544973254204 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.039455678313970566 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03518277034163475 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0450424924492836 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0413665845990181 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03707186132669449 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04196147620677948 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03897027671337128 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.034498605877161026 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04108581319451332 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03857039660215378 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03402695432305336 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04149297997355461 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03905273601412773 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03446587920188904 } ] }, "transformer_blocks.13.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07360547035932541, "diagnostic_nrmse": 0.07294099032878876, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08855222910642624, "block-scale-search": 0.08305513858795166, "gptq": 0.07360547035932541 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09233524650335312 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08631407469511032 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0767865851521492 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0893758162856102 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08338284492492676 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07399354875087738 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08855222910642624 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08305513858795166 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07360547035932541 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08952372521162033 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0841582715511322 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07442208379507065 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09163481742143631 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08647658675909042 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07622170448303223 } ] }, "transformer_blocks.13.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.104622483253479, "diagnostic_nrmse": 0.10542336851358414, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.13555556535720825, "block-scale-search": 0.12720850110054016, "gptq": 0.104622483253479 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.13555556535720825 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12720850110054016 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.104622483253479 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.13807560503482819 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12931667268276215 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.10653525590896606 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.1363499015569687 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12766779959201813 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.10520680248737335 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.13642434775829315 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1276945322751999 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.10522660613059998 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.13659340143203735 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12805026769638062 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.10535228252410889 } ] }, "transformer_blocks.13.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.058403369039297104, "diagnostic_nrmse": 0.05746734142303467, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06950840353965759, "block-scale-search": 0.06516372412443161, "gptq": 0.058403369039297104 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06950840353965759 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06516372412443161 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.058403369039297104 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07044176757335663 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06613773107528687 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.059213168919086456 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07020178437232971 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06598622351884842 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.058939605951309204 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07080958783626556 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06661080569028854 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05942480266094208 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07212864607572556 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06798626482486725 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06050381064414978 } ] }, "transformer_blocks.13.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04978174343705177, "diagnostic_nrmse": 0.04875538498163223, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06162445992231369, "block-scale-search": 0.05661327764391899, "gptq": 0.04978174343705177 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06162445992231369 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05661327764391899 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04978174343705177 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06803824752569199 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0625593513250351 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.055607520043849945 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06414102762937546 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.059302449226379395 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.052316274493932724 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06374350935220718 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05931772291660309 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05220393091440201 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06658389419317245 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06231166049838066 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.054727211594581604 } ] }, "transformer_blocks.13.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.059178903698921204, "diagnostic_nrmse": 0.05836201086640358, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07048249244689941, "block-scale-search": 0.06609774380922318, "gptq": 0.059178903698921204 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07048249244689941 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06609774380922318 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.059178903698921204 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07199757546186447 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0674893707036972 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.060351669788360596 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07140637189149857 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06706009060144424 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.059880055487155914 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0717489942908287 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06747882068157196 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06016672030091286 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07300782203674316 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0687296912074089 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.061132293194532394 } ] }, "transformer_blocks.14.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03242802247405052, "diagnostic_nrmse": 0.032456833869218826, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03912680223584175, "block-scale-search": 0.03668789193034172, "gptq": 0.03242802247405052 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0404818058013916 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.037492286413908005 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03337741643190384 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04305468872189522 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03954802826046944 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03541794791817665 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03987490013241768 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03712548688054085 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0329347588121891 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03912680223584175 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03668789193034172 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03242802247405052 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.039847638458013535 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.037520427256822586 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.033061590045690536 } ] }, "transformer_blocks.14.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.034763436764478683, "diagnostic_nrmse": 0.03423440828919411, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.042016297578811646, "block-scale-search": 0.03936047852039337, "gptq": 0.034763436764478683 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.043580301105976105 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04022567346692085 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03582771494984627 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.046006739139556885 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04229003190994263 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03790832310914993 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.042842842638492584 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.039826229214668274 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03527427092194557 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.042016297578811646 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03936047852039337 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.034763436764478683 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04251480475068092 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.040026433765888214 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.035256966948509216 } ] }, "transformer_blocks.14.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0743478536605835, "diagnostic_nrmse": 0.07430101186037064, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08950644731521606, "block-scale-search": 0.08399312943220139, "gptq": 0.0743478536605835 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09313255548477173 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08716132491827011 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07756323367357254 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08992492407560349 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08399517089128494 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07439545542001724 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08950644731521606 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08399312943220139 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0743478536605835 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09042125940322876 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08518842607736588 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07510801404714584 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09285274147987366 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0876145213842392 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07720836997032166 } ] }, "transformer_blocks.14.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.09686977416276932, "diagnostic_nrmse": 0.09837265312671661, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.12567606568336487, "block-scale-search": 0.11787982285022736, "gptq": 0.09686977416276932 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.12567606568336487 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11787982285022736 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.09686977416276932 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.1272309571504593 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11924321204423904 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.09776989370584488 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.1259172111749649 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11805540323257446 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.09696818143129349 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.12595748901367188 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11812913417816162 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.09697244316339493 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.1266491413116455 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11858267337083817 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.09731289744377136 } ] }, "transformer_blocks.14.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05493904650211334, "diagnostic_nrmse": 0.05440678820014, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06556303054094315, "block-scale-search": 0.06149514764547348, "gptq": 0.05493904650211334 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06556303054094315 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06149514764547348 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05493904650211334 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06648946553468704 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06242762878537178 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05574566498398781 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06608561426401138 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.062207795679569244 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05545112118124962 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06647844612598419 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06266220659017563 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05575863644480705 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06759127229452133 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0637161135673523 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0566033273935318 } ] }, "transformer_blocks.14.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04929477348923683, "diagnostic_nrmse": 0.04813232272863388, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06197650358080864, "block-scale-search": 0.05660179257392883, "gptq": 0.04929477348923683 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06197650358080864 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05660179257392883 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04929477348923683 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06760288029909134 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06199325993657112 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.054569389671087265 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06381528824567795 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0588369220495224 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.051414791494607925 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0636039525270462 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05912322551012039 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0514906644821167 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0669415146112442 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0624409094452858 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05428924411535263 } ] }, "transformer_blocks.14.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05512251332402229, "diagnostic_nrmse": 0.05467784404754639, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06584321707487106, "block-scale-search": 0.061742544174194336, "gptq": 0.05512251332402229 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06584321707487106 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.061742544174194336 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05512251332402229 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06760689616203308 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0635228306055069 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05663491040468216 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06668412685394287 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06274981796741486 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05590372160077095 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06678996235132217 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06288771331310272 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05593698471784592 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06764569878578186 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06374099105596542 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05660141631960869 } ] }, "transformer_blocks.15.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03355555608868599, "diagnostic_nrmse": 0.033429019153118134, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04080449789762497, "block-scale-search": 0.03830859810113907, "gptq": 0.03355555608868599 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.042436424642801285 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03921446204185486 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03470712527632713 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.044688377529382706 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04114646092057228 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.036599818617105484 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.041600923985242844 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03869276121258736 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03404267877340317 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04080449789762497 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03830859810113907 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03355555608868599 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.041594527661800385 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03912340849637985 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03422633931040764 } ] }, "transformer_blocks.15.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.034769050776958466, "diagnostic_nrmse": 0.034311000257730484, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.042407598346471786, "block-scale-search": 0.03968530148267746, "gptq": 0.034769050776958466 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04418394714593887 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04069310799241066 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03601621463894844 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0459352545440197 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04222619906067848 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03753790259361267 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04299942031502724 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03998985514044762 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.035136692225933075 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.042407598346471786 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03968530148267746 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.034769050776958466 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0429752916097641 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04036674275994301 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03530411049723625 } ] }, "transformer_blocks.15.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07287673652172089, "diagnostic_nrmse": 0.0728527307510376, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08835828304290771, "block-scale-search": 0.08287008851766586, "gptq": 0.07287673652172089 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0922626182436943 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08618482202291489 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07611539214849472 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08906076848506927 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08316951245069504 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07320379465818405 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08835828304290771 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08287008851766586 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07287673652172089 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08943074941635132 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08404453098773956 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07373140752315521 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09204090386629105 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08691500872373581 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07598469406366348 } ] }, "transformer_blocks.15.attn.to_out.0": { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0971912369132042, "diagnostic_nrmse": 0.098712258040905, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.1269211620092392, "block-scale-search": 0.11889035999774933, "gptq": 0.0971912369132042 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.13003741204738617 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12163848429918289 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.09936793893575668 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.13127204775810242 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12285573035478592 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.10043853521347046 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.12847068905830383 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12049861997365952 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0983702540397644 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.1273462176322937 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1194455698132515 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.09755630791187286 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.1269211620092392 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.11889035999774933 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0971912369132042 } ] }, "transformer_blocks.15.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05084732174873352, "diagnostic_nrmse": 0.05048166960477829, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06085939705371857, "block-scale-search": 0.05711689218878746, "gptq": 0.05084732174873352 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06085939705371857 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05711689218878746 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05084732174873352 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.062101129442453384 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05834941193461418 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.051898784935474396 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06153963878750801 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05791480466723442 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05144720524549484 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06187715753912926 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05826021358370781 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05168695002794266 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06271423399448395 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.059049736708402634 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05229587480425835 } ] }, "transformer_blocks.15.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05049499496817589, "diagnostic_nrmse": 0.04946468397974968, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06386099010705948, "block-scale-search": 0.05794459581375122, "gptq": 0.05049499496817589 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06386099010705948 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05794459581375122 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05049499496817589 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07078658044338226 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06429741531610489 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05666089057922363 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0659865289926529 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06059796363115311 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05287645384669304 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06532090157270432 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0606171190738678 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05272945761680603 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0683150514960289 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06376083940267563 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05543459579348564 } ] }, "transformer_blocks.15.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.052779655903577805, "diagnostic_nrmse": 0.05237184464931488, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06330142170190811, "block-scale-search": 0.05932234600186348, "gptq": 0.052779655903577805 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06330142170190811 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05932234600186348 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.052779655903577805 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06535548716783524 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.061435211449861526 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.054549310356378555 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06431140750646591 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06050514429807663 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05367301404476166 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06423553824424744 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06045865640044212 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0536157563328743 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06493847817182541 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06115968897938728 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0541425496339798 } ] }, "transformer_blocks.16.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03317870572209358, "diagnostic_nrmse": 0.033094968646764755, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04059768468141556, "block-scale-search": 0.038024310022592545, "gptq": 0.03317870572209358 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0426635779440403 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03939833492040634 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.034759338945150375 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0448332317173481 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0412035696208477 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03654548153281212 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04156070575118065 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.038511428982019424 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03372332081198692 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04059768468141556 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.038024310022592545 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03317870572209358 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.041048385202884674 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03863861411809921 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.033664725720882416 } ] }, "transformer_blocks.16.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0322171226143837, "diagnostic_nrmse": 0.0321250855922699, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.039453405886888504, "block-scale-search": 0.03697076812386513, "gptq": 0.0322171226143837 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.041756581515073776 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03832482919096947 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.033764638006687164 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.043548114597797394 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.039808474481105804 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03522758558392525 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04038149490952492 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03739291802048683 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0327502004802227 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.039453405886888504 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03697076812386513 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0322171226143837 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03982216864824295 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03748601675033569 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0326702855527401 } ] }, "transformer_blocks.16.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07334647327661514, "diagnostic_nrmse": 0.0732276514172554, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08945465087890625, "block-scale-search": 0.0840214416384697, "gptq": 0.07334647327661514 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09444870799779892 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08832516521215439 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07781798392534256 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0900629535317421 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08421681821346283 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07366789132356644 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08945465087890625 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0840214416384697 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07334647327661514 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09046231955289841 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08504873514175415 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07423193752765656 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0926339402794838 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08733610808849335 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07606872916221619 } ] }, "transformer_blocks.16.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.10080887377262115, "diagnostic_nrmse": 0.1010168269276619, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.1306617558002472, "block-scale-search": 0.1224767193198204, "gptq": 0.10080887377262115 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.1311977356672287 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1230957955121994 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.10133489966392517 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.13270795345306396 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12455029040575027 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.10258279740810394 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.1307443082332611 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1225641742348671 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.10107412934303284 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.1306617558002472 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1224767193198204 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.10080887377262115 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.13124673068523407 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.12308572977781296 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.10130402445793152 } ] }, "transformer_blocks.16.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04790845885872841, "diagnostic_nrmse": 0.047728531062603, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0573568232357502, "block-scale-search": 0.05384219065308571, "gptq": 0.04790845885872841 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0573568232357502 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05384219065308571 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04790845885872841 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.058609671890735626 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05510883033275604 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.048910338431596756 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05801377445459366 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05463920906186104 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04845670238137245 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.058298259973526 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05489427596330643 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04862329363822937 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.05896248668432236 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05551786720752716 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.049082718789577484 } ] }, "transformer_blocks.16.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05120106786489487, "diagnostic_nrmse": 0.05067234858870506, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06497098505496979, "block-scale-search": 0.05856438726186752, "gptq": 0.05120106786489487 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06497098505496979 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05856438726186752 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05120106786489487 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07059413194656372 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06378912925720215 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05620652064681053 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06599252671003342 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.060385122895240784 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.052689142525196075 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06565064936876297 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06072742119431496 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05278521403670311 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06896239519119263 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06441962718963623 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05588604509830475 } ] }, "transformer_blocks.16.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.049513548612594604, "diagnostic_nrmse": 0.049320120364427567, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.059430889785289764, "block-scale-search": 0.0556899718940258, "gptq": 0.049513548612594604 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.059430889785289764 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0556899718940258 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.049513548612594604 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06138497591018677 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.057659003883600235 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05117105320096016 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0602576918900013 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0566583052277565 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05019492283463478 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.060196250677108765 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05665925517678261 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05014593526721001 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.060823410749435425 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05725162476301193 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.050601474940776825 } ] }, "transformer_blocks.17.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03158959001302719, "diagnostic_nrmse": 0.031676966696977615, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03901536017656326, "block-scale-search": 0.03649554029107094, "gptq": 0.03158959001302719 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04172089695930481 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03826921060681343 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.033545318990945816 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04380455240607262 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03992782533168793 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.035216767340898514 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.040116798132658005 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03710039705038071 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03223977982997894 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03901536017656326 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03649554029107094 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03158959001302719 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.039335135370492935 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03705241158604622 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03198658302426338 } ] }, "transformer_blocks.17.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.027185967192053795, "diagnostic_nrmse": 0.027130426838994026, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03358551859855652, "block-scale-search": 0.03140273690223694, "gptq": 0.027185967192053795 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03613237664103508 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03292829543352127 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.02887582592666149 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03929542377591133 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03526530787348747 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.031264565885066986 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03494150936603546 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.032159652560949326 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.027981458231806755 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03358551859855652 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03140273690223694 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.027185967192053795 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03375697135925293 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03174668177962303 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02747790515422821 } ] }, "transformer_blocks.17.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07331082224845886, "diagnostic_nrmse": 0.07270781695842743, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09011098742485046, "block-scale-search": 0.08488446474075317, "gptq": 0.07331082224845886 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09513064473867416 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0889667421579361 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07783080637454987 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09158619493246078 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08618064224720001 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07450181245803833 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09011098742485046 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08488446474075317 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07331082224845886 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0901704952120781 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0849861204624176 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07342362403869629 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09144914150238037 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08635696768760681 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07443855702877045 } ] }, "transformer_blocks.17.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.08347146958112717, "diagnostic_nrmse": 0.0837123915553093, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.10670217871665955, "block-scale-search": 0.10023295134305954, "gptq": 0.08347146958112717 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10794580727815628 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.10121733695268631 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08402027934789658 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.10844121873378754 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.10191110521554947 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.08472620695829391 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.10698752105236053 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1005050390958786 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.08363722264766693 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.10670217871665955 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.10023295134305954 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.08347146958112717 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10687648504972458 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.10043573379516602 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0836360976099968 } ] }, "transformer_blocks.17.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04941168054938316, "diagnostic_nrmse": 0.049346908926963806, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.059562381356954575, "block-scale-search": 0.0559161938726902, "gptq": 0.04941168054938316 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.059562381356954575 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0559161938726902 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04941168054938316 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06094558164477348 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.057264115661382675 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05051928386092186 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06030028685927391 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05668012052774429 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04994629696011543 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.060652267187833786 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.057046372443437576 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.050260115414857864 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06136767566204071 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05767883360385895 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0507170595228672 } ] }, "transformer_blocks.17.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.043577004224061966, "diagnostic_nrmse": 0.04247526824474335, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.055522333830595016, "block-scale-search": 0.05041979253292084, "gptq": 0.043577004224061966 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.055522333830595016 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05041979253292084 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.043577004224061966 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0616733580827713 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05614659935235977 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0492328479886055 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05755942314863205 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05296554043889046 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.045885708183050156 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.05742943286895752 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05339360237121582 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.046040166169404984 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06111224740743637 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05717245861887932 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04905840381979942 } ] }, "transformer_blocks.17.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04838436096906662, "diagnostic_nrmse": 0.04832955077290535, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05848614498972893, "block-scale-search": 0.05479242280125618, "gptq": 0.04838436096906662 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.05848614498972893 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05479242280125618 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04838436096906662 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.060308173298835754 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05654206871986389 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04990927875041962 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0592673160135746 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05570194497704506 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04905392602086067 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.05937369167804718 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.055751413106918335 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.049074895679950714 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0600263811647892 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05645018070936203 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.049612972885370255 } ] }, "transformer_blocks.18.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03276408091187477, "diagnostic_nrmse": 0.03285849466919899, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.040291931480169296, "block-scale-search": 0.037696585059165955, "gptq": 0.03276408091187477 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.043732866644859314 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03999380022287369 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.035271160304546356 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.045195262879133224 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0412350557744503 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.036469120532274246 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04143434017896652 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0383208692073822 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03349565342068672 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.040291931480169296 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.037696585059165955 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03276408091187477 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04065415635704994 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03820282593369484 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03315277770161629 } ] }, "transformer_blocks.18.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.029812041670084, "diagnostic_nrmse": 0.029750391840934753, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.036752477288246155, "block-scale-search": 0.03427949547767639, "gptq": 0.029812041670084 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04010307416319847 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.036396574229002 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.032094795256853104 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04309874400496483 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03871066868305206 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03440074250102043 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.038193605840206146 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.035073172301054 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.030663534998893738 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.036752477288246155 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03427949547767639 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.029812041670084 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.036827802658081055 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.034648992121219635 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03010226972401142 } ] }, "transformer_blocks.18.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07730743288993835, "diagnostic_nrmse": 0.07701003551483154, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09441239386796951, "block-scale-search": 0.08874817937612534, "gptq": 0.07730743288993835 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10313680768013 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09647893905639648 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08502298593521118 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09466025233268738 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08862385898828506 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07734993100166321 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09441239386796951 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08874817937612534 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07730743288993835 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09566421806812286 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09003972262144089 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07824844866991043 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09776440262794495 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09203965961933136 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07992003113031387 } ] }, "transformer_blocks.18.attn.to_out.0": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0811941847205162, "diagnostic_nrmse": 0.08161794394254684, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.1053021103143692, "block-scale-search": 0.09849075973033905, "gptq": 0.0811941847205162 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.1053021103143692 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09849075973033905 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0811941847205162 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.10671231895685196 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09987899661064148 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.08218517899513245 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.10559976100921631 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09905749559402466 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.08152974396944046 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.10596011579036713 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09957225620746613 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.08190423995256424 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10704967379570007 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.1005079597234726 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.08255083858966827 } ] }, "transformer_blocks.18.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.048985812813043594, "diagnostic_nrmse": 0.048837531358003616, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.059627700597047806, "block-scale-search": 0.05598372593522072, "gptq": 0.048985812813043594 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.059627700597047806 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05598372593522072 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.048985812813043594 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06076102331280708 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.057139765471220016 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04993809759616852 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.060262780636548996 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.056655772030353546 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0494283102452755 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.060902319848537445 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.057223837822675705 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.049886979162693024 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06210313364863396 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05839760601520538 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05073738470673561 } ] }, "transformer_blocks.18.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04863281548023224, "diagnostic_nrmse": 0.046790264546871185, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06315211206674576, "block-scale-search": 0.056937139481306076, "gptq": 0.04863281548023224 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06315211206674576 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.056937139481306076 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04863281548023224 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0692257359623909 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06256083399057388 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05409792438149452 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06425195187330246 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.058837391436100006 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05025948956608772 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06457702070474625 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.060126204043626785 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05105498433113098 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06989273428916931 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0656895637512207 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05523448437452316 } ] }, "transformer_blocks.18.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.048094820231199265, "diagnostic_nrmse": 0.047923725098371506, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.058683380484580994, "block-scale-search": 0.05497770383954048, "gptq": 0.048094820231199265 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.058683380484580994 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05497770383954048 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.048094820231199265 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.060153648257255554 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05638695880770683 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04931957274675369 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05939411371946335 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05578339844942093 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04871927946805954 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.059803687036037445 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05616408959031105 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0489593967795372 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06092773377895355 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05724059417843819 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04972467198967934 } ] }, "transformer_blocks.19.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.029176006093621254, "diagnostic_nrmse": 0.028999963775277138, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03621744364500046, "block-scale-search": 0.03387779742479324, "gptq": 0.029176006093621254 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03880048915743828 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.035495124757289886 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0310384389013052 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.041285738348960876 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.037366028875112534 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.032935746014118195 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03723570331931114 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.034330036491155624 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.029783958569169044 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03621744364500046 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03387779742479324 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.029176006093621254 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03673110902309418 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.034547701478004456 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02966384030878544 } ] }, "transformer_blocks.19.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.027480587363243103, "diagnostic_nrmse": 0.027271829545497894, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03428822010755539, "block-scale-search": 0.031879499554634094, "gptq": 0.027480587363243103 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03655266389250755 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.033255789428949356 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.029066206887364388 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04450317472219467 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03920598328113556 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03486967831850052 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.036919306963682175 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03346841037273407 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.029234033077955246 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03428822010755539 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.031879499554634094 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.027480587363243103 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03429502621293068 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0322205014526844 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02772345021367073 } ] }, "transformer_blocks.19.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07787445187568665, "diagnostic_nrmse": 0.07809273898601532, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09617787599563599, "block-scale-search": 0.09045572578907013, "gptq": 0.07787445187568665 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10408979654312134 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0973871648311615 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08495642244815826 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09659602493047714 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09040416032075882 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07800832390785217 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09617787599563599 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09045572578907013 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07787445187568665 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09759530425071716 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0918143168091774 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07901906222105026 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10011450946331024 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09444267302751541 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.08108697831630707 } ] }, "transformer_blocks.19.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.08177726715803146, "diagnostic_nrmse": 0.08224071562290192, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.10441148281097412, "block-scale-search": 0.09748649597167969, "gptq": 0.08177726715803146 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10832233726978302 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.10030937194824219 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08430866152048111 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.10882099717855453 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.10081150382757187 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.08452775329351425 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.10543103516101837 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09799765795469284 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.08205166459083557 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.10441148281097412 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09748649597167969 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.08177726715803146 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10436544567346573 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09787813574075699 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.08187201619148254 } ] }, "transformer_blocks.19.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0489770770072937, "diagnostic_nrmse": 0.04862327128648758, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.059826578944921494, "block-scale-search": 0.05608179047703743, "gptq": 0.0489770770072937 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.059826578944921494 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05608179047703743 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0489770770072937 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.061276089400053024 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0574922040104866 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.050174273550510406 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06069646775722504 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.057087432593107224 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04970264434814453 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06149975210428238 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.057836104184389114 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.050217948853969574 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06302810460329056 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05928131937980652 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05124637484550476 } ] }, "transformer_blocks.19.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05579201877117157, "diagnostic_nrmse": 0.05404052138328552, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0731525719165802, "block-scale-search": 0.06612631678581238, "gptq": 0.05579201877117157 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0731525719165802 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06612631678581238 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05579201877117157 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07746604830026627 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07084580510854721 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.060294948518276215 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07361093163490295 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06822869926691055 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05736943334341049 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.075686514377594 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07112904638051987 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.059550683945417404 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08341020345687866 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07907910645008087 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06518479436635971 } ] }, "transformer_blocks.19.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04724839702248573, "diagnostic_nrmse": 0.046925704926252365, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05777124688029289, "block-scale-search": 0.054137200117111206, "gptq": 0.04724839702248573 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.05777124688029289 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.054137200117111206 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04724839702248573 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.059440821409225464 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05574318394064903 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04859013110399246 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05875197425484657 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.055223528295755386 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.048037488013505936 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.059347376227378845 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05576721951365471 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04841530695557594 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06068481504917145 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05697941780090332 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.049288176000118256 } ] }, "transformer_blocks.20.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.030711688101291656, "diagnostic_nrmse": 0.03027183748781681, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03832068294286728, "block-scale-search": 0.035724300891160965, "gptq": 0.030711688101291656 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.041286103427410126 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.037545088678598404 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03288233280181885 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.042031094431877136 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.038234893232584 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03344660997390747 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.038993850350379944 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.036029696464538574 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.031113114207983017 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03832068294286728 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.035724300891160965 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.030711688101291656 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03919554501771927 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.036743100732564926 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03146705776453018 } ] }, "transformer_blocks.20.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.029172195121645927, "diagnostic_nrmse": 0.028856266289949417, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.036564406007528305, "block-scale-search": 0.03393245488405228, "gptq": 0.029172195121645927 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.039481133222579956 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03557617589831352 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03110084868967533 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04330860450863838 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.038599852472543716 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03409034386277199 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03809672221541405 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03474633768200874 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.030085112899541855 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.036564406007528305 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03393245488405228 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.029172195121645927 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.036902640014886856 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03458201885223389 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.029605865478515625 } ] }, "transformer_blocks.20.attn.to_v": { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07488932460546494, "diagnostic_nrmse": 0.07513297349214554, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09322547167539597, "block-scale-search": 0.08701695501804352, "gptq": 0.07488932460546494 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10279180854558945 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09567664563655853 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08348074555397034 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09322547167539597 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08701695501804352 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07488932460546494 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09336137026548386 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08755232393741608 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07519190013408661 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09570375084877014 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08986233174800873 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07706455886363983 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09960725903511047 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09355345368385315 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07997376471757889 } ] }, "transformer_blocks.20.attn.to_out.0": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07294793426990509, "diagnostic_nrmse": 0.07329118996858597, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09578823298215866, "block-scale-search": 0.08988708257675171, "gptq": 0.07294793426990509 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09640708565711975 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08991172909736633 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07296629250049591 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0965048149228096 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09034416079521179 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07332850992679596 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09578823298215866 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08988708257675171 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07294793426990509 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09661602228879929 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09059620648622513 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07395654171705246 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09800349175930023 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09238370507955551 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07508485019207001 } ] }, "transformer_blocks.20.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04537477716803551, "diagnostic_nrmse": 0.044662512838840485, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05576789006590843, "block-scale-search": 0.05233735218644142, "gptq": 0.04537477716803551 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.05576789006590843 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05233735218644142 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04537477716803551 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05748295038938522 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05391472205519676 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04672859236598015 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.056805700063705444 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05333426967263222 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04606135934591293 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.057669952511787415 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05421308055520058 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.046735405921936035 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.05929170548915863 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0557817704975605 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04775666072964668 } ] }, "transformer_blocks.20.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.060149285942316055, "diagnostic_nrmse": 0.057108074426651, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08033174276351929, "block-scale-search": 0.07228695601224899, "gptq": 0.060149285942316055 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08033174276351929 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07228695601224899 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.060149285942316055 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0826231986284256 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0753658339381218 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06304829567670822 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07943441718816757 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07395660877227783 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06119435280561447 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08329033106565475 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07873611897230148 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06469187885522842 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09554678201675415 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09063491970300674 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07299903780221939 } ] }, "transformer_blocks.20.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0418059267103672, "diagnostic_nrmse": 0.04105880483984947, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.051469411700963974, "block-scale-search": 0.048244476318359375, "gptq": 0.0418059267103672 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.051469411700963974 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.048244476318359375 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0418059267103672 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05215902999043465 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04884395748376846 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04224807396531105 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05193452909588814 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04874199256300926 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.042101528495550156 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.052948590368032455 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04968726634979248 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.042860016226768494 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0546744205057621 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05133424326777458 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04399263858795166 } ] }, "transformer_blocks.21.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.025405418127775192, "diagnostic_nrmse": 0.024714887142181396, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03190719336271286, "block-scale-search": 0.029733523726463318, "gptq": 0.025405418127775192 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03516308218240738 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.031623732298612595 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.02761016972362995 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03500247001647949 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.031525131314992905 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0273763258010149 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0324040912091732 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.029832808300852776 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0255927424877882 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03190719336271286 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.029733523726463318 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.025405418127775192 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03256744146347046 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0305892676115036 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.025969691574573517 } ] }, "transformer_blocks.21.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.022833917289972305, "diagnostic_nrmse": 0.022413033992052078, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.028738312423229218, "block-scale-search": 0.02672073058784008, "gptq": 0.022833917289972305 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03163950890302658 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028204243630170822 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.02466348558664322 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03615298494696617 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03141842037439346 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.02768191136419773 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.030629925429821014 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.027708733454346657 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.023973604664206505 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.028738312423229218 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02672073058784008 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.022833917289972305 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.028993116691708565 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.027178971096873283 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.023093707859516144 } ] }, "transformer_blocks.21.attn.to_v": { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07593726366758347, "diagnostic_nrmse": 0.0755472183227539, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09517958760261536, "block-scale-search": 0.0891365110874176, "gptq": 0.07593726366758347 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10740762203931808 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09983377903699875 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08661945164203644 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09517958760261536 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0891365110874176 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07593726366758347 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09593035280704498 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08997133374214172 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07651564478874207 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09827537089586258 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09236375242471695 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07838866859674454 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10225658863782883 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09618675708770752 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0812615230679512 } ] }, "transformer_blocks.21.attn.to_out.0": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07476174831390381, "diagnostic_nrmse": 0.07495087385177612, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09628883749246597, "block-scale-search": 0.09025808423757553, "gptq": 0.07476174831390381 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09695863723754883 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09045504033565521 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07514901459217072 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09717497229576111 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09102147072553635 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07535891234874725 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09628883749246597 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09025808423757553 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07476174831390381 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09612036496400833 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09037113189697266 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07492081820964813 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09662303328514099 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09057893604040146 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07511206716299057 } ] }, "transformer_blocks.21.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.045065682381391525, "diagnostic_nrmse": 0.04389350116252899, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0557556115090847, "block-scale-search": 0.052266817539930344, "gptq": 0.045065682381391525 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0557556115090847 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.052266817539930344 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.045065682381391525 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05763212591409683 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05400585010647774 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.046570658683776855 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05684120953083038 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05340910702943802 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04581759497523308 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.057640355080366135 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.054254259914159775 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04641666263341904 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.059539299458265305 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.055971402674913406 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.047631245106458664 } ] }, "transformer_blocks.21.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06709925085306168, "diagnostic_nrmse": 0.06560405343770981, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08934496343135834, "block-scale-search": 0.080225370824337, "gptq": 0.06709925085306168 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08934496343135834 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.080225370824337 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06709925085306168 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09066062420606613 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08287042379379272 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06944100558757782 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08799704909324646 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08251093327999115 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0679398626089096 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09354516118764877 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08843100816011429 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07310519367456436 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10671678185462952 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.10182201117277145 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.08198820054531097 } ] }, "transformer_blocks.21.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04147197678685188, "diagnostic_nrmse": 0.04032756760716438, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05142440274357796, "block-scale-search": 0.04807233065366745, "gptq": 0.04147197678685188 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.05142440274357796 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04807233065366745 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04147197678685188 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05323652923107147 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04979705810546875 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04295334592461586 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.052355989813804626 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.049120672047138214 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04218777269124985 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.05313991755247116 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04987674579024315 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04270093888044357 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.05482615530490875 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05153224989771843 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.043839626014232635 } ] }, "transformer_blocks.22.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.02510991133749485, "diagnostic_nrmse": 0.024404237046837807, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03165540099143982, "block-scale-search": 0.02936977706849575, "gptq": 0.02510991133749485 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03473852574825287 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03115452639758587 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.027294769883155823 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.038356274366378784 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.033949051052331924 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.030012642964720726 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03324832767248154 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03030897118151188 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.026177627965807915 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03165540099143982 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02936977706849575 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.02510991133749485 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03196080029010773 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.029939549043774605 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.025467410683631897 } ] }, "transformer_blocks.22.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.020208626985549927, "diagnostic_nrmse": 0.019590971991419792, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.02536321058869362, "block-scale-search": 0.02354765683412552, "gptq": 0.020208626985549927 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.027751430869102478 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.024888113141059875 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.021844832226634026 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03312303498387337 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028587862849235535 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.025390807539224625 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.027187583968043327 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02461893856525421 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.02139885351061821 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.02536321058869362 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02354765683412552 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.020208626985549927 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.025514619424939156 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.023940548300743103 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02041883021593094 } ] }, "transformer_blocks.22.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07152161002159119, "diagnostic_nrmse": 0.07077808678150177, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08975141495466232, "block-scale-search": 0.0840647965669632, "gptq": 0.07152161002159119 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09912747889757156 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09219886362552643 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08021816611289978 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09007539600133896 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08399192988872528 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07170595973730087 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08975141495466232 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0840647965669632 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07152161002159119 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09136433154344559 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08565570414066315 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0728200301527977 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09464038908481598 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08865537494421005 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07518384605646133 } ] }, "transformer_blocks.22.attn.to_out.0": { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07860570400953293, "diagnostic_nrmse": 0.07859751582145691, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.10047346353530884, "block-scale-search": 0.09437084197998047, "gptq": 0.07860570400953293 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10478426516056061 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09795389324426651 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08199328929185867 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.10438164323568344 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09766925871372223 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.08183185756206512 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.10192859917879105 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09558115899562836 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07991884648799896 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.10057654231786728 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09436899423599243 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0787864401936531 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10047346353530884 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09437084197998047 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07860570400953293 } ] }, "transformer_blocks.22.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04785647243261337, "diagnostic_nrmse": 0.0463755838572979, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05931134149432182, "block-scale-search": 0.05545695871114731, "gptq": 0.04785647243261337 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.05931134149432182 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05545695871114731 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04785647243261337 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06162086874246597 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05773218348622322 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.049680259078741074 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06038026511669159 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05671078711748123 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04857916384935379 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06089315935969353 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05719699338078499 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04883334040641785 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06233852356672287 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05857183784246445 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04970816895365715 } ] }, "transformer_blocks.22.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07515943050384521, "diagnostic_nrmse": 0.07393142580986023, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.10027393698692322, "block-scale-search": 0.08967616409063339, "gptq": 0.07515943050384521 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10027393698692322 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08967616409063339 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07515943050384521 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.10114042460918427 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0910264179110527 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0765412300825119 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09837740659713745 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09077616780996323 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07528747618198395 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.10294398665428162 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09639570862054825 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07943205535411835 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.11497663706541061 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.10864647477865219 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.08808314800262451 } ] }, "transformer_blocks.22.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.042845677584409714, "diagnostic_nrmse": 0.04139789193868637, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05321792885661125, "block-scale-search": 0.0496516078710556, "gptq": 0.042845677584409714 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.05321792885661125 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0496516078710556 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.042845677584409714 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05462263152003288 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0509917289018631 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04397246614098549 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05378177389502525 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05034051463007927 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.043223924934864044 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.05437294393777847 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.050997599959373474 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.043604858219623566 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.05596829950809479 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.052536193281412125 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.044604189693927765 } ] }, "transformer_blocks.23.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.024578826501965523, "diagnostic_nrmse": 0.023641018196940422, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03094867803156376, "block-scale-search": 0.028650877997279167, "gptq": 0.024578826501965523 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.034221500158309937 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03053230233490467 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.026830404996871948 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03798457235097885 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03329510614275932 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.029539983719587326 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.032450806349515915 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02955186925828457 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.025585487484931946 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03094867803156376 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028650877997279167 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.024578826501965523 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.031269315630197525 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02926911972463131 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.024926766753196716 } ] }, "transformer_blocks.23.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.01833047717809677, "diagnostic_nrmse": 0.017745425924658775, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.022937966510653496, "block-scale-search": 0.02128405123949051, "gptq": 0.01833047717809677 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.02542021870613098 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.022590065374970436 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.01990300603210926 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.029945168644189835 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.025595057755708694 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.022761965170502663 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.024590322747826576 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.022213425487279892 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.01936987414956093 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.022937966510653496 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02128405123949051 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.01833047717809677 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.023131774738430977 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02169114723801613 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.01855095662176609 } ] }, "transformer_blocks.23.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06997082382440567, "diagnostic_nrmse": 0.06841670721769333, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08795422315597534, "block-scale-search": 0.08217068016529083, "gptq": 0.06997082382440567 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09652527421712875 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08940809965133667 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0780394896864891 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08877948671579361 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08268708735704422 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07059235870838165 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08795422315597534 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08217068016529083 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06997082382440567 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08910945057868958 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0833778902888298 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07092254608869553 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0920206680893898 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08622351288795471 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07305478304624557 } ] }, "transformer_blocks.23.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0615641213953495, "diagnostic_nrmse": 0.061659373342990875, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07789678871631622, "block-scale-search": 0.07324475049972534, "gptq": 0.0615641213953495 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07883498817682266 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07393229007720947 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.062235891819000244 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07851967215538025 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07370563596487045 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06210210919380188 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07793296873569489 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07328451424837112 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06165085360407829 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07789678871631622 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07324475049972534 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0615641213953495 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07841792702674866 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07375914603471756 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06194658577442169 } ] }, "transformer_blocks.23.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.052350182086229324, "diagnostic_nrmse": 0.0506369024515152, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06495027989149094, "block-scale-search": 0.060731738805770874, "gptq": 0.052350182086229324 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06495027989149094 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.060731738805770874 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.052350182086229324 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06726275384426117 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06296855956315994 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05413157865405083 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06619013100862503 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.062064625322818756 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05313699319958687 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06693946570158005 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06286866962909698 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05368078127503395 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06887013465166092 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06466423720121384 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05485793203115463 } ] }, "transformer_blocks.23.img_mlp.out": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06882953643798828, "diagnostic_nrmse": 0.06973747164011002, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09068700671195984, "block-scale-search": 0.08287166804075241, "gptq": 0.06882953643798828 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09250380098819733 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08282022923231125 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0696306824684143 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09435652941465378 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08462969213724136 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07109162956476212 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09068700671195984 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08287166804075241 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06882953643798828 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0939866453409195 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08760268241167068 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07218008488416672 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10477571934461594 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09928140044212341 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0800146833062172 } ] }, "transformer_blocks.23.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04377076029777527, "diagnostic_nrmse": 0.042190615087747574, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.05434572324156761, "block-scale-search": 0.05070468410849571, "gptq": 0.04377076029777527 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.05434572324156761 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05070468410849571 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04377076029777527 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05555164813995361 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05181826651096344 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0445990264415741 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0549018569290638 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05137449502944946 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04404182359576225 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.05590008199214935 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05238547921180725 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.044733352959156036 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.05773404240608215 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05417044833302498 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04595595598220825 } ] }, "transformer_blocks.24.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.022947434335947037, "diagnostic_nrmse": 0.02207670919597149, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.02877461351454258, "block-scale-search": 0.026676597073674202, "gptq": 0.022947434335947037 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03196341544389725 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028313253074884415 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.024928614497184753 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03676610440015793 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.031757306307554245 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.028293803334236145 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0305468887090683 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02767849527299404 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.024110695347189903 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.02877461351454258 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.026676597073674202 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.022947434335947037 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.02898087538778782 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.027158575132489204 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.023201093077659607 } ] }, "transformer_blocks.24.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.01670953631401062, "diagnostic_nrmse": 0.016063034534454346, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.020792854949831963, "block-scale-search": 0.01932443678379059, "gptq": 0.01670953631401062 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.023150181397795677 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.020531857386231422 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.018122993409633636 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.027630791068077087 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02352851815521717 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.020977843552827835 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.022486725822091103 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.020259158685803413 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.01774175465106964 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.020792854949831963 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.01932443678379059 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.01670953631401062 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.02092023938894272 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.019621996209025383 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.016860786825418472 } ] }, "transformer_blocks.24.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05687319487333298, "diagnostic_nrmse": 0.055229220539331436, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07172340154647827, "block-scale-search": 0.06663095951080322, "gptq": 0.05687319487333298 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07675878703594208 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07115902751684189 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06231147050857544 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07317967712879181 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06747543811798096 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.057842858135700226 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07172340154647827 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06663095951080322 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05687319487333298 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07204025983810425 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06742673367261887 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05742795765399933 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07374517619609833 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06908668577671051 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.058605875819921494 } ] }, "transformer_blocks.24.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05576368793845177, "diagnostic_nrmse": 0.056278426200151443, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07092501223087311, "block-scale-search": 0.06667263060808182, "gptq": 0.05576368793845177 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07167119532823563 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0673636943101883 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05638131499290466 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07184858620166779 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06751988083124161 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.056472957134246826 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07101649791002274 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06674507260322571 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.055850885808467865 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07092501223087311 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06667263060808182 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05576368793845177 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07151883095502853 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06732255965471268 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05619257315993309 } ] }, "transformer_blocks.24.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05447795242071152, "diagnostic_nrmse": 0.05328325927257538, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06754114478826523, "block-scale-search": 0.06308729201555252, "gptq": 0.05447795242071152 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06754114478826523 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06308729201555252 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05447795242071152 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06971632689237595 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0651579424738884 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.056174781173467636 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06817563623189926 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0638880580663681 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05478401482105255 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06886164844036102 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06462087482213974 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05522480234503746 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07058363407850266 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06619657576084137 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05628885328769684 } ] }, "transformer_blocks.24.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06824438273906708, "diagnostic_nrmse": 0.06715560704469681, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08933509886264801, "block-scale-search": 0.07991626113653183, "gptq": 0.06824438273906708 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08933509886264801 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07991626113653183 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06824438273906708 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09326809644699097 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0832589790225029 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07127095013856888 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08901119977235794 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08117420226335526 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06865876168012619 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09204873442649841 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08575352281332016 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07196146249771118 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10177888721227646 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09616955369710922 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07959356904029846 } ] }, "transformer_blocks.24.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04824547842144966, "diagnostic_nrmse": 0.04717031121253967, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.059914637356996536, "block-scale-search": 0.05590759217739105, "gptq": 0.04824547842144966 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.059914637356996536 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05590759217739105 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04824547842144966 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.061313532292842865 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05709804594516754 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04916113615036011 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06027334928512573 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05642126128077507 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.048357974737882614 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0610954724252224 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05716375261545181 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04893261566758156 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06269598752260208 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05873589590191841 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04994577541947365 } ] }, "transformer_blocks.25.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.021411092951893806, "diagnostic_nrmse": 0.02063668705523014, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.026749461889266968, "block-scale-search": 0.024760402739048004, "gptq": 0.021411092951893806 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.02965603955090046 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.026357389986515045 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.023304296657443047 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0339168980717659 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.029363518580794334 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.026265941560268402 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.028210239484906197 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02556035667657852 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.022334029898047447 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.026749461889266968 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.024760402739048004 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.021411092951893806 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.027117948979139328 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.025425970554351807 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02179831638932228 } ] }, "transformer_blocks.25.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.01585524156689644, "diagnostic_nrmse": 0.015305365435779095, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.019646205008029938, "block-scale-search": 0.018253207206726074, "gptq": 0.01585524156689644 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.021862151101231575 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.019364718347787857 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.017194963991642 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.02539568394422531 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.021843712776899338 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.019581498578190804 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.020964404568076134 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.018933188170194626 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.01663942262530327 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.019646205008029938 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.018253207206726074 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.01585524156689644 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.01990201510488987 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.018670672550797462 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.016109637916088104 } ] }, "transformer_blocks.25.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05871327966451645, "diagnostic_nrmse": 0.05693086236715317, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07346959412097931, "block-scale-search": 0.06853605806827545, "gptq": 0.05871327966451645 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08082804083824158 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07504585385322571 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06599726527929306 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07392387092113495 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06848654896020889 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05888903886079788 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07346959412097931 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06853605806827545 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05871327966451645 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07484430074691772 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06990483403205872 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05990462005138397 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07773512601852417 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0728948786854744 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06204579398036003 } ] }, "transformer_blocks.25.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06839209794998169, "diagnostic_nrmse": 0.0685078501701355, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08633605390787125, "block-scale-search": 0.08126381784677505, "gptq": 0.06839209794998169 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08779361099004745 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08248043805360794 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06955520808696747 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08733293414115906 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08204422891139984 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06905031204223633 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08632619678974152 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08122379332780838 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06839588284492493 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08633605390787125 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08126381784677505 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06839209794998169 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0870213657617569 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08192506432533264 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06889207661151886 } ] }, "transformer_blocks.25.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06409569084644318, "diagnostic_nrmse": 0.06299937516450882, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0789307951927185, "block-scale-search": 0.07383477687835693, "gptq": 0.06409569084644318 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0789307951927185 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07383477687835693 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06409569084644318 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0809989795088768 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07567084580659866 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06553919613361359 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07961064577102661 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07457529753446579 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0643683448433876 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08043917268514633 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0754576325416565 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06492376327514648 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08269699662923813 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07748254388570786 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06629236042499542 } ] }, "transformer_blocks.25.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06431746482849121, "diagnostic_nrmse": 0.061956677585840225, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08316159248352051, "block-scale-search": 0.07538853585720062, "gptq": 0.06431746482849121 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08316159248352051 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07538853585720062 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06431746482849121 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08651085942983627 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07879305630922318 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06751632690429688 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08389419317245483 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0782594382762909 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06620011478662491 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08843609690666199 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0834258496761322 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07050461322069168 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10012824833393097 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0955224260687828 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07915288209915161 } ] }, "transformer_blocks.25.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05750753730535507, "diagnostic_nrmse": 0.05661112442612648, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07098101824522018, "block-scale-search": 0.06623488664627075, "gptq": 0.05750753730535507 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07098101824522018 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06623488664627075 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05750753730535507 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07356584817171097 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0685715302824974 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05946063622832298 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07187028974294662 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06711708754301071 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.057931676506996155 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07229894399642944 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06767221540212631 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05827353149652481 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07426049560308456 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06952551007270813 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.059506312012672424 } ] }, "transformer_blocks.26.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.030426738783717155, "diagnostic_nrmse": 0.029586538672447205, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03772810846567154, "block-scale-search": 0.03504718095064163, "gptq": 0.030426738783717155 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04199695959687233 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03751673921942711 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03326959162950516 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04696749150753021 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04137337580323219 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03711371123790741 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03986567631363869 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03623224049806595 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03174208849668503 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03772810846567154 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03504718095064163 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.030426738783717155 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03817690163850784 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03574904054403305 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.030888458713889122 } ] }, "transformer_blocks.26.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.02197456546127796, "diagnostic_nrmse": 0.02139921858906746, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.027242757380008698, "block-scale-search": 0.025246405974030495, "gptq": 0.02197456546127796 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.030320031568408012 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.026922937482595444 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.023933399468660355 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03558969870209694 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03044559806585312 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.027287958189845085 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.029240116477012634 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.026341091841459274 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.02319057658314705 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.027242757380008698 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.025246405974030495 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.02197456546127796 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.027386706322431564 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02563219703733921 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02221764251589775 } ] }, "transformer_blocks.26.attn.to_v": { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0751919150352478, "diagnostic_nrmse": 0.07462207227945328, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09269094467163086, "block-scale-search": 0.08658912032842636, "gptq": 0.0751919150352478 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10545632243156433 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09804932773113251 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08664879947900772 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09269094467163086 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08658912032842636 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0751919150352478 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09303461015224457 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08719508349895477 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07563071697950363 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09539689123630524 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08940029889345169 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07735022902488708 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09971589595079422 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09362030029296875 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.08071816712617874 } ] }, "transformer_blocks.26.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07897636294364929, "diagnostic_nrmse": 0.07899927347898483, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09978499263525009, "block-scale-search": 0.09373491257429123, "gptq": 0.07897636294364929 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10114916414022446 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09471844881772995 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08009632676839828 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.10092920064926147 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09477531909942627 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.08008943498134613 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09994340687990189 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09375819563865662 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07913088798522949 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09978499263525009 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09373491257429123 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07897636294364929 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.1003405973315239 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09426857531070709 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07927994430065155 } ] }, "transformer_blocks.26.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06449618190526962, "diagnostic_nrmse": 0.063340924680233, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07921389490365982, "block-scale-search": 0.07395052909851074, "gptq": 0.06449618190526962 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07921389490365982 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07395052909851074 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06449618190526962 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08181294798851013 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07647500187158585 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06657441705465317 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08023685961961746 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07507596164941788 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06508749723434448 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08098194003105164 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07585864514112473 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06558961421251297 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08296585828065872 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0777941644191742 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06685776263475418 } ] }, "transformer_blocks.26.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.059803299605846405, "diagnostic_nrmse": 0.05548469349741936, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07788011431694031, "block-scale-search": 0.07086487114429474, "gptq": 0.059803299605846405 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07788011431694031 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07086487114429474 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.059803299605846405 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08177081495523453 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07496413588523865 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06368418037891388 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07831704616546631 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0734262764453888 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.061346229165792465 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08315860480070114 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07882377505302429 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06534451246261597 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09442716836929321 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09110287576913834 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07309830188751221 } ] }, "transformer_blocks.26.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05375339090824127, "diagnostic_nrmse": 0.052713874727487564, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06620381027460098, "block-scale-search": 0.061670877039432526, "gptq": 0.05375339090824127 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06620381027460098 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.061670877039432526 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05375339090824127 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06827452778816223 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06349565833806992 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.055289510637521744 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06708300113677979 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06263251602649689 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05430382862687111 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06773151457309723 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06333859264850616 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05474359914660454 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06951117515563965 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06499674916267395 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05591997131705284 } ] }, "transformer_blocks.27.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.024782465770840645, "diagnostic_nrmse": 0.02402728982269764, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03078167326748371, "block-scale-search": 0.02854415774345398, "gptq": 0.024782465770840645 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03413883224129677 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.030422141775488853 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.026975661516189575 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03807251900434494 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.033380404114723206 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.029815662652254105 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03235689178109169 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02944200672209263 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0257975272834301 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03078167326748371 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02854415774345398 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.024782465770840645 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.031082600355148315 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02912023849785328 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02517419494688511 } ] }, "transformer_blocks.27.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.015389843843877316, "diagnostic_nrmse": 0.01491739135235548, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.01892334781587124, "block-scale-search": 0.017562154680490494, "gptq": 0.015389843843877316 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.020923268049955368 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.018645619973540306 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.01658780314028263 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.025973929092288017 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.022009417414665222 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.01980382390320301 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.020741749554872513 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.018583497032523155 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.016466505825519562 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.01892334781587124 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.017562154680490494 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.015389843843877316 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.018909625709056854 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.01774367317557335 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.015444271266460419 } ] }, "transformer_blocks.27.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06363964080810547, "diagnostic_nrmse": 0.06171553209424019, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07874297350645065, "block-scale-search": 0.07366321980953217, "gptq": 0.06363964080810547 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0859326496720314 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07995419204235077 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07060806453227997 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08018266409635544 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07446137070655823 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.064479298889637 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07874297350645065 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07366321980953217 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06363964080810547 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07960356771945953 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07464008778333664 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06439698487520218 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08272584527730942 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0777839943766594 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06679287552833557 } ] }, "transformer_blocks.27.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07498054206371307, "diagnostic_nrmse": 0.07487048953771591, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09425992518663406, "block-scale-search": 0.08861158788204193, "gptq": 0.07498054206371307 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0952959954738617 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08929513394832611 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07550870627164841 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09538957476615906 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08947895467281342 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07583583146333694 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09436605125665665 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08853399753570557 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07505116611719131 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09425992518663406 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08861158788204193 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07498054206371307 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0948527604341507 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08915073424577713 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07545286417007446 } ] }, "transformer_blocks.27.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06707865744829178, "diagnostic_nrmse": 0.06650357693433762, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08279482275247574, "block-scale-search": 0.07728487998247147, "gptq": 0.06707865744829178 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08279482275247574 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07728487998247147 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06707865744829178 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08578035980463028 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0800468698143959 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06938987225294113 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08370288461446762 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07842134684324265 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06763140112161636 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08423589915037155 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07896783947944641 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0679100900888443 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08617842197418213 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08083794265985489 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06920939683914185 } ] }, "transformer_blocks.27.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.043281152844429016, "diagnostic_nrmse": 0.03608354926109314, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.057080551981925964, "block-scale-search": 0.05196435749530792, "gptq": 0.043281152844429016 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.057080551981925964 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05196435749530792 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.043281152844429016 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.060376666486263275 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05573888495564461 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.046625666320323944 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.05788969248533249 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05426906421780586 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04489351063966751 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.060861553996801376 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05824102461338043 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.04765196144580841 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06907358765602112 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06729016453027725 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05327899008989334 } ] }, "transformer_blocks.27.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05799639970064163, "diagnostic_nrmse": 0.05739786848425865, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.071836456656456, "block-scale-search": 0.06688615679740906, "gptq": 0.05799639970064163 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.071836456656456 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06688615679740906 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05799639970064163 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07528544962406158 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06994988769292831 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06084117665886879 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07286985963582993 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06792407482862473 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05868102237582207 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07301557064056396 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06822391599416733 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.058699339628219604 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07469461858272552 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06991277635097504 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.05984991788864136 } ] }, "transformer_blocks.28.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.034289728850126266, "diagnostic_nrmse": 0.033401258289813995, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04273591563105583, "block-scale-search": 0.039731573313474655, "gptq": 0.034289728850126266 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04677015170454979 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.042395006865262985 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03735603764653206 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05059262737631798 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04534511640667915 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04030538722872734 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04436386749148369 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.040659770369529724 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.035331662744283676 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04273591563105583 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.039731573313474655 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.034289728850126266 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04336028918623924 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.040663767606019974 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0349874421954155 } ] }, "transformer_blocks.28.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.026747101917862892, "diagnostic_nrmse": 0.026053009554743767, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03331863135099411, "block-scale-search": 0.030983850359916687, "gptq": 0.026747101917862892 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.03678658977150917 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03297782316803932 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.029050596058368683 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04173724725842476 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.036519769579172134 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0325658842921257 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03529053181409836 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03200972452759743 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.027939818799495697 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03331863135099411 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.030983850359916687 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.026747101917862892 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.033590346574783325 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03148236125707626 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.027080198749899864 } ] }, "transformer_blocks.28.attn.to_v": { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07403647154569626, "diagnostic_nrmse": 0.07350705564022064, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0916408821940422, "block-scale-search": 0.08571582287549973, "gptq": 0.07403647154569626 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.10170476138591766 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09465236216783524 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.08312542736530304 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0916408821940422 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08571582287549973 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07403647154569626 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09182039648294449 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08601529151201248 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07423966377973557 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0936562716960907 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08794228732585907 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07567355036735535 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0970141589641571 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09120024740695953 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0781361311674118 } ] }, "transformer_blocks.28.attn.to_out.0": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07157959043979645, "diagnostic_nrmse": 0.07024399936199188, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0940302386879921, "block-scale-search": 0.08787697553634644, "gptq": 0.07157959043979645 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0947851836681366 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08837562054395676 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0717935860157013 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09458719193935394 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08852693438529968 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07214707881212234 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0940302386879921 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08787697553634644 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.07157959043979645 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09504377841949463 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08901138603687286 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07253847271203995 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09694741666316986 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.09073252975940704 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07355731725692749 } ] }, "transformer_blocks.28.img_mlp.proj": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06812631338834763, "diagnostic_nrmse": 0.06735935062170029, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08448225259780884, "block-scale-search": 0.07893206179141998, "gptq": 0.06812631338834763 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08448225259780884 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07893206179141998 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06812631338834763 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08702483773231506 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08122696727514267 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.07007893174886703 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08542602509260178 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07992357015609741 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06861938536167145 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08622203767299652 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0807364284992218 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06913511455059052 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0883357897400856 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08283035457134247 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07063183188438416 } ] }, "transformer_blocks.28.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04989105463027954, "diagnostic_nrmse": 0.043820805847644806, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06627953052520752, "block-scale-search": 0.06044197827577591, "gptq": 0.04989105463027954 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06627953052520752 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06044197827577591 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04989105463027954 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07007596641778946 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0645381361246109 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05357600748538971 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06765790283679962 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06351868063211441 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0522555373609066 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07075047492980957 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0675501599907875 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.055244993418455124 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.07893583178520203 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07625290006399155 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06052577868103981 } ] }, "transformer_blocks.28.img_mlp.gate_layer": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.056967753916978836, "diagnostic_nrmse": 0.05636787414550781, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.07094636559486389, "block-scale-search": 0.06605125218629837, "gptq": 0.056967753916978836 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.07094636559486389 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06605125218629837 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.056967753916978836 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.07291821390390396 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06786001473665237 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05848443880677223 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.07159171253442764 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06686219573020935 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.057372502982616425 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.07202713191509247 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06740346550941467 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05766581371426582 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0737835168838501 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06905153393745422 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.058900490403175354 } ] }, "transformer_blocks.29.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.027843372896313667, "diagnostic_nrmse": 0.027159642428159714, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0346028096973896, "block-scale-search": 0.03231354430317879, "gptq": 0.027843372896313667 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.037834085524082184 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03427845239639282 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.030087493360042572 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.041645169258117676 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03719508275389671 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.032956190407276154 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03628772869706154 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03324795514345169 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.02885950170457363 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0346028096973896 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03231354430317879 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.027843372896313667 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.03498787432909012 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03285980597138405 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02824576385319233 } ] }, "transformer_blocks.29.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.019777562469244003, "diagnostic_nrmse": 0.01937846839427948, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.024561118334531784, "block-scale-search": 0.022850675508379936, "gptq": 0.019777562469244003 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.026834972202777863 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02415868267416954 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.02126591093838215 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03303375467658043 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028545435518026352 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.02563389576971531 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.026591463014483452 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.024036429822444916 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.021030860021710396 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.024561118334531784 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.022850675508379936 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.019777562469244003 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.024556465446949005 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.023138871416449547 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.019932614639401436 } ] }, "transformer_blocks.29.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06886471807956696, "diagnostic_nrmse": 0.06892576068639755, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08531203120946884, "block-scale-search": 0.08016038686037064, "gptq": 0.06886471807956696 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09274189919233322 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08663904666900635 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07589223235845566 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08606452494859695 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08039984107017517 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06932014226913452 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08531203120946884 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08016038686037064 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06886471807956696 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0863414779305458 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08122619241476059 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06972650438547134 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08876737952232361 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08377793431282043 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07178196310997009 } ] }, "transformer_blocks.29.attn.to_out.0": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.071787029504776, "diagnostic_nrmse": 0.07160255312919617, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09530378878116608, "block-scale-search": 0.08928383886814117, "gptq": 0.071787029504776 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09715040028095245 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0909733921289444 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0729844719171524 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09577500075101852 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08999604731798172 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0722360610961914 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.09530378878116608 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08928383886814117 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.071787029504776 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09529662132263184 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08950378000736237 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07195848226547241 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.09684886783361435 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0908118262887001 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0728888213634491 } ] }, "transformer_blocks.29.img_mlp.proj": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06574791669845581, "diagnostic_nrmse": 0.06505517661571503, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08244342356920242, "block-scale-search": 0.0771661102771759, "gptq": 0.06574791669845581 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08239446580410004 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07692720741033554 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06591872870922089 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08377246558666229 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07830110192298889 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0669596940279007 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08244342356920242 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0771661102771759 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06574791669845581 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08297739923000336 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07784510403871536 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06617695093154907 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0847010612487793 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07954816520214081 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06750665605068207 } ] }, "transformer_blocks.29.img_mlp.out": { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06827475875616074, "diagnostic_nrmse": 0.06430605053901672, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.09128279238939285, "block-scale-search": 0.08246040344238281, "gptq": 0.06827475875616074 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.09128279238939285 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08246040344238281 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06827475875616074 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09267541766166687 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08411084115505219 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06951901316642761 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08977391570806503 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08312667161226273 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06834417581558228 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.09276477992534637 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0869985967874527 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.07149119675159454 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.10063780099153519 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0957174226641655 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.07714049518108368 } ] }, "transformer_blocks.29.img_mlp.gate_layer": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05078013986349106, "diagnostic_nrmse": 0.05045003443956375, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06381142139434814, "block-scale-search": 0.05956286936998367, "gptq": 0.05078013986349106 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06388683617115021 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05948587507009506 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05096594616770744 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06512726843357086 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06066027656197548 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05190344154834747 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06381142139434814 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05956286936998367 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05078013986349106 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06398171186447144 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05990375205874443 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.050952158868312836 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.06526579707860947 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06120101362466812 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.051937684416770935 } ] }, "transformer_blocks.30.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.029307911172509193, "diagnostic_nrmse": 0.028718452900648117, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.03654775395989418, "block-scale-search": 0.03420555964112282, "gptq": 0.029307911172509193 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0391789972782135 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03574134781956673 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.031147684901952744 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.043445270508527756 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03938518092036247 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03479127585887909 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03807684779167175 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.035048238933086395 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.030326589941978455 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.03654775395989418 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.03420555964112282 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.029307911172509193 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.036833979189395905 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.034693241119384766 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.029644589871168137 } ] }, "transformer_blocks.30.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.022792652249336243, "diagnostic_nrmse": 0.022499531507492065, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.028438657522201538, "block-scale-search": 0.02659420296549797, "gptq": 0.022792652249336243 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.030695220455527306 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.027866527438163757 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.02428710274398327 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03447328507900238 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.030745258554816246 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.02717067301273346 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.029742008075118065 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.027323659509420395 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.023639138787984848 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.028438657522201538 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02659420296549797 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.022792652249336243 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.028566772118210793 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.02694706991314888 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.02304719015955925 } ] }, "transformer_blocks.30.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06591947376728058, "diagnostic_nrmse": 0.06565987318754196, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0821121335029602, "block-scale-search": 0.07707247138023376, "gptq": 0.06591947376728058 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08807620406150818 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08242636173963547 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.07165919989347458 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.0829249769449234 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07740362733602524 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06631084531545639 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0821121335029602 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07707247138023376 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06591947376728058 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08286429196596146 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07811838388442993 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06672435998916626 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08498132973909378 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08036220818758011 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06848935782909393 } ] }, "transformer_blocks.30.attn.to_out.0": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06744065135717392, "diagnostic_nrmse": 0.06734875589609146, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08909554779529572, "block-scale-search": 0.08397991955280304, "gptq": 0.06744065135717392 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0937555730342865 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08694075047969818 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06979308277368546 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.09113550931215286 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08507616817951202 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06857573986053467 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08998945355415344 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08397924900054932 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0678626298904419 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08909554779529572 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08397991955280304 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06744065135717392 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08958917111158371 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0844993144273758 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06779250502586365 } ] }, "transformer_blocks.30.img_mlp.proj": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06586595624685287, "diagnostic_nrmse": 0.0653093010187149, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08418480306863785, "block-scale-search": 0.07870837301015854, "gptq": 0.06586595624685287 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08551505953073502 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07941622287034988 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06679347157478333 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08582180738449097 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07994716614484787 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06720039248466492 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08418480306863785 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07870837301015854 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.06586595624685287 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.08413073420524597 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07899251580238342 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0660354271531105 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.08535732328891754 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.08043792843818665 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.06716299057006836 } ] }, "transformer_blocks.30.img_mlp.out": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.062071334570646286, "diagnostic_nrmse": 0.05973590910434723, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.08189795166254044, "block-scale-search": 0.07603129744529724, "gptq": 0.062071334570646286 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.08464843779802322 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07646451890468597 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.06291095912456512 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.08385427296161652 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07647705078125 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.06309441477060318 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.08189795166254044 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.07603129744529724 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.062071334570646286 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.0847616195678711 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0797223374247551 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.06502513587474823 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0925607979297638 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0878763273358345 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.0711161345243454 } ] }, "transformer_blocks.30.img_mlp.gate_layer": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05095790699124336, "diagnostic_nrmse": 0.050940293818712234, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.06503831595182419, "block-scale-search": 0.06100768968462944, "gptq": 0.05095790699124336 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06657630205154419 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06157519295811653 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.05189099535346031 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.06721170246601105 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.062277503311634064 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.05254780128598213 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.06535114347934723 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06105396896600723 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.05112536624073982 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.06503831595182419 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.06100768968462944 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.05095790699124336 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.0659777820110321 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.061991605907678604 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.051734261214733124 } ] }, "transformer_blocks.31.attn.to_q": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0234739501029253, "diagnostic_nrmse": 0.022979263216257095, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.029933325946331024, "block-scale-search": 0.028059223666787148, "gptq": 0.0234739501029253 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.0318174734711647 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028945008292794228 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.024625524878501892 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.03382774814963341 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.030763894319534302 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.02645184099674225 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.03085438348352909 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028482487425208092 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.02399245835840702 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.029933325946331024 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028059223666787148 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.0234739501029253 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.030431684106588364 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.028737910091876984 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.023967312648892403 } ] }, "transformer_blocks.31.attn.to_k": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.015668785199522972, "diagnostic_nrmse": 0.015626437962055206, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.019858017563819885, "block-scale-search": 0.01860971190035343, "gptq": 0.015668785199522972 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.021207137033343315 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.019258778542280197 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.01647157594561577 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.024463120847940445 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.021764326840639114 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.018946312367916107 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.021004334092140198 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.01924181915819645 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.016366368159651756 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.019858017563819885 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.01860971190035343 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.015668785199522972 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.02001565508544445 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.01888938620686531 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.015884213149547577 } ] }, "transformer_blocks.31.attn.to_v": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04452911391854286, "diagnostic_nrmse": 0.044074174016714096, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.056988030672073364, "block-scale-search": 0.05338676646351814, "gptq": 0.04452911391854286 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.06105947121977806 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.056772299110889435 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.04791877046227455 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05806364491581917 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05385108292102814 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04503055661916733 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.056988030672073364 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05338676646351814 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04452911391854286 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.05747601389884949 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05422360822558403 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.045052286237478256 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.05917296186089516 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.056094422936439514 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04663001373410225 } ] }, "transformer_blocks.31.attn.to_out.0": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0350814163684845, "diagnostic_nrmse": 0.03490816429257393, "rank": 128, "shape": [ 4096, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.04779575765132904, "block-scale-search": 0.04504045844078064, "gptq": 0.0350814163684845 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.04962616413831711 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04676062986254692 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.036358024924993515 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.04819902777671814 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04523535072803497 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.03524860367178917 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.04779575765132904 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04504045844078064 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0350814163684845 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04815934598445892 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04531922936439514 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.03541629761457443 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04866310954093933 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04573095589876175 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.03561161831021309 } ] }, "transformer_blocks.31.img_mlp.proj": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03563684597611427, "diagnostic_nrmse": 0.035137731581926346, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.0471821203827858, "block-scale-search": 0.04399576410651207, "gptq": 0.03563684597611427 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.049891673028469086 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04558117687702179 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.03727889060974121 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.048712633550167084 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04493331164121628 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0366080068051815 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.0471821203827858 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04399576410651207 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.03563684597611427 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.04698735848069191 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.044296082109212875 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.035720065236091614 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.04777441918849945 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.04511096701025963 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.036420296877622604 } ] }, "transformer_blocks.31.img_mlp.out": { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.005825801752507687, "diagnostic_nrmse": 0.0060274358838796616, "rank": 128, "shape": [ 4096, 12288 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.007076249923557043, "block-scale-search": 0.006694824434816837, "gptq": 0.005825801752507687 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.011302146129310131 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.0076413825154304504 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.006829113699495792 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.010063925758004189 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.007541584316641092 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.0066779605112969875 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.00792334508150816 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.007021376397460699 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.0061205909587442875 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.007076249923557043 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.006694824434816837 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.005825801752507687 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.007508125156164169 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.007213526405394077 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.006179172080010176 } ] }, "transformer_blocks.31.img_mlp.gate_layer": { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04212893173098564, "diagnostic_nrmse": 0.04142777621746063, "rank": 128, "shape": [ 12288, 4096 ], "sample_rows": 2048, "fit_rows": 512, "diagnostic_rows": 512, "hessian_tokens": 3623738, "same_alpha_fit_nrmse_by_rounding": { "plain": 0.055843010544776917, "block-scale-search": 0.052149686962366104, "gptq": 0.04212893173098564 }, "candidates": [ { "smoothing_alpha": 0.0, "weight_rounding": "plain", "fit_nrmse": 0.059946704655885696 }, { "smoothing_alpha": 0.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.054599329829216 }, { "smoothing_alpha": 0.0, "weight_rounding": "gptq", "fit_nrmse": 0.0446600615978241 }, { "smoothing_alpha": 0.25, "weight_rounding": "plain", "fit_nrmse": 0.05697661638259888 }, { "smoothing_alpha": 0.25, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05273302271962166 }, { "smoothing_alpha": 0.25, "weight_rounding": "gptq", "fit_nrmse": 0.04277932643890381 }, { "smoothing_alpha": 0.5, "weight_rounding": "plain", "fit_nrmse": 0.055843010544776917 }, { "smoothing_alpha": 0.5, "weight_rounding": "block-scale-search", "fit_nrmse": 0.052149686962366104 }, { "smoothing_alpha": 0.5, "weight_rounding": "gptq", "fit_nrmse": 0.04212893173098564 }, { "smoothing_alpha": 0.75, "weight_rounding": "plain", "fit_nrmse": 0.055703967809677124 }, { "smoothing_alpha": 0.75, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05229783430695534 }, { "smoothing_alpha": 0.75, "weight_rounding": "gptq", "fit_nrmse": 0.042212024331092834 }, { "smoothing_alpha": 1.0, "weight_rounding": "plain", "fit_nrmse": 0.056714773178100586 }, { "smoothing_alpha": 1.0, "weight_rounding": "block-scale-search", "fit_nrmse": 0.05340028181672096 }, { "smoothing_alpha": 1.0, "weight_rounding": "gptq", "fit_nrmse": 0.04303791746497154 } ] } }