← Back to Quantization

mixed int7/int8

Quantization
Used in
6 PRs
Best BPB
1.0579
Avg BPB
1.0597

Hyperparameters Across PRs

pr_numberbitsscope
1925—embeddings and row gate
2007—embeddings and model weights
2060—embeddings and blocks
2109—embeddings and attention gate
21578embeddings and selected groups
21637embeddings and attention gates