fix(_internals): use n_tokens0 offset when enabling last-token logits in add_sequence #3290
| Job | Run time |
|---|---|
| 32s | |
| 3m 44s | |
| 3m 22s | |
| 3m 32s | |
| 11m 42s | |
| 2m 58s | |
| 3m 16s | |
| 3m 40s | |
| 9m 2s | |
| 3m 13s | |
| 8m 12s | |
| 3m 20s | |
| 10m 37s | |
| 8m 15s | |
| 1h 15m 25s |
| Job | Run time |
|---|---|
| 32s | |
| 3m 44s | |
| 3m 22s | |
| 3m 32s | |
| 11m 42s | |
| 2m 58s | |
| 3m 16s | |
| 3m 40s | |
| 9m 2s | |
| 3m 13s | |
| 8m 12s | |
| 3m 20s | |
| 10m 37s | |
| 8m 15s | |
| 1h 15m 25s |