← Back to Quantization

mixed int8/int6

Quantization
Used in
3 PRs
Best BPB
1.0788
Avg BPB
1.2910

Submissions

PR #172by GMaN1911
1.1812
PR #435by rthgit
1.6130
PR #1718by himanshudongre
1.0788

Hyperparameters Across PRs

pr_numberbitsscope
1728attention int8, MLP int6
435—selected weights
17188control tensors, small matrices, tok_emb