← Back to Architecture
attention sink
ArchitectureUsed in
2 PRs
Best BPB
1.0738
Avg BPB
1.0762
Hyperparameters Across PRs
| pr_number | parameters |
|---|---|
| 1895 | {"sink_keys_per_kv_head":1} |
| 1980 | — |
| pr_number | parameters |
|---|---|
| 1895 | {"sink_keys_per_kv_head":1} |
| 1980 | — |