← Back to Quantization

int5 QAT

Quantization
Used in
2 PRs
Best BPB
1.1326
Avg BPB
1.3134

Submissions

PR #669by amabito
1.4942
PR #861by JoeProAI
1.1326

Hyperparameters Across PRs

pr_numberbitsscope
6695all matrix weights; embeddings remain fp16
8615all weights