donghyunli/Meta-Llama-3-70B-KronQ-W4A16

Meta-Llama-3-70B — KronQ W4A16 (packed int4)

Paper: arXiv:2607.07964 · Code: GitHub

Meta-Llama-3-70B quantized to 4-bit weights / 16-bit activations with KronQ (Kronecker-factored Hessian quantization). Weights are stored packed int4 (~36 GB); a fused dequant + bidirectional-incoherence (BiIP) CUDA kernel unpacks them on the fly.

Results (WikiText-2, seqlen 2048)

Perplexity: 3.26

Zero-shot accuracy:

PIQA	ARC-E	ARC-C	HellaSwag	WinoGrande	BoolQ	OBQA	Average
84.22	81.73	61.69	84.75	79.08	86.82	48.00	75.18

(lm-evaluation-harness, 0-shot. acc_norm for PIQA/HellaSwag/ARC/OBQA, acc for WinoGrande/BoolQ.)

Usage

KronQ-packed checkpoint. Load with the KronQ runtime:

python eval_pretrained.py meta-llama/Meta-Llama-3-70B donghyunli/Meta-Llama-3-70B-KronQ-W4A16 --ppl --zs

Recipe

Per-channel asymmetric W4, weight-only (a_bits=16), --alpha 0.25, BiIP, act_order, raw H_G. 128 WikiText-2 calibration sequences.

License

Derivative of Meta-Llama-3-70B — llama3 license.