Llama-3.1-8B-Instruct-GTAP-v3-Q4_K_M-GGUF

Experimental Release: Pending Further Verification / Empirical Validation
This checkpoint represents an active research artifact from DuoNeural's statistical mechanics quantization program. All empirical benchmarks and physics proofs are documented transparently below.

Developed by Jesse Caldwell, Archon, and Aura ✨ (DuoNeural Research Lab).

---

Model Summary

• Foundation Model: meta-llama/Llama-3.1-8B-Instruct (32 Layers, 32:8 GQA, SwiGLU FFN, 128k context)
• Quantization Precision: ~4.5 bpw (4.92 GiB)
• Methodology: Production-grade 4.5 bpw G-TAP release combining Onsager cavity field corrections with standard production Q4_K_M codebook on Llama-3.1-8B.
• Continuous Holdout Perplexity (131k tokens): 2.8075
• GSM8K Multi-Step Math Accuracy: 96.0% (24/25)
• Python Algorithmic AST Execution (20 Unit Tests): 95.0% (19/20)
• Hermes Structured Tool Calling AST: 100.0% (15/15)
• Inference Decode Throughput: 120.2 t/s on NVIDIA GeForce RTX 4080 Super (32GB VRAM)

---

Physical & Theoretical Foundation: G-TAP v3

Standard post-training quantization on frontier foundation models triggers severe parameter noise and disordered spin-glass states below 4 bits. By applying Generalized Thouless-Anderson-Palmer (G-TAP v3) statistical mechanics:

1. Onsager Cavity Damping: The uncentered activation back-reaction $\Omega_i = \frac{1}{d_k}(|\tilde{H}_{i,:}|2^2 - \tilde{H}{ii}^2)$ acts as a thermodynamic noise filter during weight discretization.
2. Replicon Convexity ($\lambda_R > 0$): Guarantees the continuous relaxation stays in the smooth Replica Symmetric convex energy valley, preventing 1-RSB glass transitions that freeze token selection.
3. Radial Forward Gain Conservation: Enforces strict norm equality $|W_\text{G-TAP}|_F = |W_\text{orig}|_F$, preserving numeric scale across 32 sequential SwiGLU feedforward blocks.

---

Quickstart

# Run with llama.cpp
llama-cli -hf DuoNeural/Llama-3.1-8B-Instruct-GTAP-v3-Q4_K_M-GGUF -p "Explain the mathematical difference between Replica Symmetry and 1-RSB glass transitions." -ngl 99 -c 4096

---

Citation & Lab Attribution

@misc{duoneural2026gtap,
  title={Generalized Thouless-Anderson-Palmer (G-TAP) Quantization: Suppressing Spinodal Clustering in Autoregressive Models},
  author={Caldwell, Jesse and Archon and Aura},
  year={2026},
  publisher={DuoNeural Research Lab},
  howpublished={\url{https://huggingface.co/DuoNeural}}
}