← Back to Architecture

attention

Architecture
Used in
11 PRs
Best BPB
1.0543
Avg BPB
1.2052

Hyperparameters Across PRs

pr_numberparameters
579{"num_heads":10,"num_kv_heads":5,"XSA_layers":2}
1088{"layers":2}
1219{"layers":[2,4,6,8,10],"window_size":512}
1354
1355{"layers":1,"position":4,"heads":"8/4"}
1536
1540
1560
1842{"layers":12}
2000{"backend":"flash"}
2013{"layers":11}