Meta-Llama-3-8B — KronQ W4A16 g128 (packed int4)
Paper: arXiv:2607.07964 · Code: GitHub
Meta-Llama-3-8B quantized to 4-bit weights / 16-bit activations, group size 128, with KronQ (Kronecker-factored Hessian quantization). Packed int4 (~5.5 GB); a fused dequant + bidirectional-incoherence (BiIP) CUDA kernel with per-group scales unpacks on the fly.
Results (WikiText-2, seqlen 2048)
Perplexity: 6.337
Zero-shot accuracy:
| PIQA | ARC-E | ARC-C | HellaSwag | WinoGrande | BoolQ | OBQA | Average |
|---|---|---|---|---|---|---|---|
| 80.36 | 78.87 | 53.24 | 78.37 | 73.32 | 80.92 | 45.80 | 70.13 |
(lm-evaluation-harness 0-shot; acc_norm for PIQA/HellaSwag/ARC, acc for WinoGrande/BoolQ.)
Usage
python eval_pretrained.py meta-llama/Meta-Llama-3-8B donghyunli/Meta-Llama-3-8B-KronQ-W4A16-g128 --ppl --zs
Recipe
Group-128 asymmetric W4, weight-only (a_bits=16), --alpha 0.25, --act_order, BiIP, raw H_G. 128 WikiText-2 calibration sequences.
License
Derivative of Meta-Llama-3-8B — llama3 license.