← Back to Quantization
STE QAT (late QAT) + Full GPTQ + Int5 MLP re-quantization + GPTQ-lite
QuantizationUsed in
1 PRs
Best BPB
1.1418
Avg BPB
1.1418
Submissions
Hyperparameters Across PRs
| pr_number | bits | scope |
|---|---|---|
| 601 | 6 | all linear layers with special Int5 re-quantization for MLP |