← Back to Quantization

mixed int6/fp16

Quantization
Used in
1 PRs
Best BPB
1.1632
Avg BPB
1.1632

Submissions

PR #66by arjun-krishna1
1.1632

Hyperparameters Across PRs

pr_numberbitsscope
666MLP and attention weights int6, tied embedding fp16 passthrough