← Back to Quantization

mixed int6/int7

Quantization
Used in
14 PRs
Best BPB
1.0135
Avg BPB
1.0585

Hyperparameters Across PRs

pr_numberbitsscope
1707all
1934embeddings
1953weights + embeddings
1958model weights
19926attention/MLP matrices
19927token embeddings
1995matrix weights + embeddings
20676weights and embeddings
21206weights and embeddings
2123embeddings and block weights
2124embeddings and block weights
2128attn/MLP int6, embeddings int7
2130model weights
2133weights + embeddings