← Back to Quantization

int6 per-row with GPTQ-lite clip search

Quantization
Used in
2 PRs
Best BPB
1.0944
Avg BPB
1.0964

Submissions

PR #628by Christopher-Lee-McClendon
1.0983
PR #644by Christopher-Lee-McClendon
1.0944

Hyperparameters Across PRs

pr_numberbitsscope
6286all model tensors including embeddings
6446all model tensors including embeddings