← Back to Quantization

int6 QAT

Quantization
Used in
15 PRs
Best BPB
0.9393
Avg BPB
1.1773

Hyperparameters Across PRs

pr_numberbitsscope
1876all
3256all
3386block weights
3906all
4036all
4066model weights
4986MLP and attention weights
4996MLP and attention weights
5376all model weights
5526
5736MLP + attention
6456all
7026all
8106all
13836model