← Back to Quantization
mixed int6/int7
QuantizationUsed in
14 PRs
Best BPB
1.0135
Avg BPB
1.0585
Submissions
PR #1707by nothingLiva
1.0740PR #1934by liujshi
1.0599PR #1953by andrewbaggio1RECORD
1.0586PR #1958by okezue
1.0135PR #1992by jamesEmerson112
1.0511PR #1992by jamesEmerson112
1.0511PR #1995by User123331
1.0878PR #2067by jiashenggu
1.0592PR #2120by newjordan
1.0624PR #2123by vaibhavmishra1
1.0593PR #2124by vaibhavmishra1
1.0593PR #2128by okezue
1.0677PR #2130by TanishGudise
1.0567PR #2133by codemath3000
1.0576Hyperparameters Across PRs
| pr_number | bits | scope |
|---|---|---|
| 1707 | — | all |
| 1934 | — | embeddings |
| 1953 | — | weights + embeddings |
| 1958 | — | model weights |
| 1992 | 6 | attention/MLP matrices |
| 1992 | 7 | token embeddings |
| 1995 | — | matrix weights + embeddings |
| 2067 | 6 | weights and embeddings |
| 2120 | 6 | weights and embeddings |
| 2123 | — | embeddings and block weights |
| 2124 | — | embeddings and block weights |
| 2128 | — | attn/MLP int6, embeddings int7 |
| 2130 | — | model weights |
| 2133 | — | weights + embeddings |