{ "@context": { "@language": "en", "@vocab": "https://schema.org/", "sc": "https://schema.org/", "cr": "http://mlcommons.org/croissant/", "rai": "http://mlcommons.org/croissant/RAI/", "prov": "http://www.w3.org/ns/prov#", "dct": "http://purl.org/dc/terms/", "arrayShape": "cr:arrayShape", "citeAs": "cr:citeAs", "column": "cr:column", "conformsTo": "dct:conformsTo", "containedIn": "cr:containedIn", "data": { "@id": "cr:data", "@type": "@json" }, "dataType": { "@id": "cr:dataType", "@type": "@vocab" }, "equivalentProperty": "cr:equivalentProperty", "examples": { "@id": "cr:examples", "@type": "@json" }, "extract": "cr:extract", "field": "cr:field", "fileProperty": "cr:fileProperty", "fileObject": "cr:fileObject", "fileSet": "cr:fileSet", "format": "cr:format", "includes": "cr:includes", "isArray": "cr:isArray", "isLiveDataset": "cr:isLiveDataset", "jsonPath": "cr:jsonPath", "key": "cr:key", "md5": "cr:md5", "parentField": "cr:parentField", "path": "cr:path", "recordSet": "cr:recordSet", "references": "cr:references", "regex": "cr:regex", "repeated": "cr:repeated", "replace": "cr:replace", "samplingRate": "cr:samplingRate", "separator": "cr:separator", "source": "cr:source", "subField": "cr:subField", "transform": "cr:transform" }, "@type": "sc:Dataset", "name": "StableHLO-Spec-30", "description": "Hand-authored NL→MLIR pairs for StableHLO dialect covering 10 op families.", "conformsTo": "http://mlcommons.org/croissant/1.1", "url": "https://huggingface.co/datasets/plawanrath/StableHLO-Spec-30", "sameAs": [ "https://github.com/plawanrath/slm-to-mlir-constrained-emitter/tree/main/eval/benchmarks/stablehlo_spec_30" ], "license": "https://spdx.org/licenses/Apache-2.0.html", "creator": { "@type": "Person", "name": "Plawan Kumar Rath", "url": "https://huggingface.co/plawanrath" }, "datePublished": "2026-04-21", "dateModified": "2026-09-25", "version": "1.0.0", "inLanguage": "en", "keywords": [ "MLIR", "StableHLO", "compiler IR", "code generation", "constrained decoding", "benchmark", "NeurIPS 2026" ], "citeAs": "@inproceedings{rath2026crossdialect,\n title = {Cross-Dialect Generalization Without Retraining: Benchmarks and Evaluation of Schema-Derived Constrained Decoding for {MLIR}},\n author = {Rath, Plawan Kumar},\n booktitle = {Advances in Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track},\n year = {2026},\n eprint = {2607.18254},\n archivePrefix = {arXiv}\n}", "isAccessibleForFree": true, "distribution": [ { "@type": "cr:FileObject", "@id": "test.jsonl", "name": "test.jsonl", "description": "All records (single test split).", "contentUrl": "https://huggingface.co/datasets/plawanrath/StableHLO-Spec-30/resolve/main/data/test.jsonl", "encodingFormat": "application/jsonlines", "sha256": "921eee53012cce78a7054fbba6f537eef2b8cb6721668e47cb8ab740d76f1272" } ], "recordSet": [ { "@type": "cr:RecordSet", "@id": "records", "name": "records", "description": "One MLIR prompt/reference pair per record.", "field": [ { "@type": "cr:Field", "@id": "records/id", "name": "id", "source": { "fileObject": { "@id": "test.jsonl" }, "extract": { "column": "id" } }, "dataType": "sc:Text", "description": "Unique record identifier." }, { "@type": "cr:Field", "@id": "records/difficulty", "name": "difficulty", "source": { "fileObject": { "@id": "test.jsonl" }, "extract": { "column": "difficulty" } }, "dataType": "sc:Text", "description": "Author-assigned difficulty or 'programmatic'." }, { "@type": "cr:Field", "@id": "records/nl", "name": "nl", "source": { "fileObject": { "@id": "test.jsonl" }, "extract": { "column": "nl" } }, "dataType": "sc:Text", "description": "Natural-language description." }, { "@type": "cr:Field", "@id": "records/mlir", "name": "mlir", "source": { "fileObject": { "@id": "test.jsonl" }, "extract": { "column": "mlir" } }, "dataType": "sc:Text", "description": "Reference MLIR that verifies under mlir-opt/iree-compile." }, { "@type": "cr:Field", "@id": "records/notes", "name": "notes", "source": { "fileObject": { "@id": "test.jsonl" }, "extract": { "column": "notes" } }, "dataType": "sc:Text" }, { "@type": "cr:Field", "@id": "records/dialect", "name": "dialect", "source": { "fileObject": { "@id": "test.jsonl" }, "extract": { "column": "dialect" } }, "dataType": "sc:Text", "description": "MLIR dialect of the reference program." } ] } ], "rai:dataCollection": "Hand-authored by the dataset author against the target MLIR dialect's ODS.", "rai:dataBiases": [ "Author-curated: prompts reflect the dataset author's mental model of the target dialect; may under-represent op combinations not present in the spec examples.", "No human-subject data; no PII; no demographic bias dimensions apply." ], "rai:dataLimitations": [ "Verify-valid pass-rate measures structural validity under mlir-opt/iree-compile, not functional correctness. Programs that pass the gate may still compute the wrong function.", "English natural-language descriptions only.", "Small n (30-200 prompts per dataset) yields CI half-widths of ~3-10pp at p=0.5." ], "rai:annotationsPerExample": 0, "rai:annotationDemographics": "N/A — no human annotators.", "rai:personalSensitiveInformation": "None.", "rai:dataUseCases": [ "Evaluating NL→MLIR generation systems (constrained or unconstrained) under a verifier-based pass-rate metric.", "Out of scope: evaluating functional correctness without an additional lowering + execution harness; training or fine-tuning production code-generation models without a separate held-out corpus." ], "rai:dataSocialImpact": "These benchmarks are intended to evaluate compiler-IR generation systems against MLIR/StableHLO verifiers. They contain no human-subject, demographic, or PII content, and pose no direct risk to individuals or groups. The principal social risk is methodological: over-reliance on verifier-pass-rate as a proxy for functional correctness could lead to overstated claims about model capability. Downstream users should pair these benchmarks with execution/property-based testing before drawing conclusions about deployment readiness.", "rai:hasSyntheticData": false, "prov:wasDerivedFrom": [ { "@id": "https://openxla.org/stablehlo/spec" } ], "prov:wasGeneratedBy": [ { "@type": "prov:Activity", "prov:label": "Data collection / authoring", "description": "Hand-authored NL/StableHLO pairs across 10 op families, written against the StableHLO specification.", "prov:wasAssociatedWith": { "@type": "prov:Agent", "prov:label": "Plawan Kumar Rath" } }, { "@type": "prov:Activity", "prov:label": "Verification", "description": "Every reference program was checked before inclusion: stablehlo-opt v1.4.0 and iree-compile --compile-to=input (iree-compiler 20241104.1068, pinned). Only verifier-clean programs are included in the release." }, { "@type": "prov:Activity", "prov:label": "Freeze and export", "description": "Source records (examples/*.json in the code repository) frozen and exported 1-to-1 to data/test.jsonl; no filtering, deduplication or annotation beyond the steps above. Details: the paper (https://arxiv.org/abs/2607.18254) and the datasheet (https://github.com/plawanrath/slm-to-mlir-constrained-emitter/blob/main/docs/datasheets/datasheet.md)." } ] }