← Back to Quantization

STE QAT (late QAT) + Full GPTQ + Int5 MLP re-quantization + GPTQ-lite

Quantization
Used in
1 PRs
Best BPB
1.1418
Avg BPB
1.1418

Hyperparameters Across PRs

pr_numberbitsscope
6016all linear layers with special Int5 re-quantization for MLP