← Back to Quantization

mixed int6/int7/int8

Quantization
Used in
4 PRs
Best BPB
1.0554
Avg BPB
1.0579

Submissions

PR #2097by OnlyJundong
1.0585
PR #2132by codemath3000
1.0576
PR #2162by uniagent-alpha
1.0603
PR #2164by vimeto
1.0554

Hyperparameters Across PRs

pr_numberbitsscope
2097—tok_emb.weight
2132—model weights
2162—weights, embeddings, attention gate
2164—mixed