transformers
Fix mask slicing for models with HybridCache
#35681
Merged

Fix mask slicing for models with HybridCache #35681

Cyrilvallez merged 16 commits into main from fix-fa2-hybrid
Cyrilvallez
Cyrilvallez correctly slice
e1247d11
Cyrilvallez Cyrilvallez requested a review from ArthurZucker ArthurZucker 1 year ago
HuggingFaceDocBuilderDev
ArthurZucker
ArthurZucker commented on 2025-01-14
Cyrilvallez check mask
9c59b354
Cyrilvallez Update modular_gemma2.py
9f47f44f
Cyrilvallez fix
ed2e3d7b
Cyrilvallez add tests
94145727
Cyrilvallez Cyrilvallez requested a review from Rocketknight1 Rocketknight1 1 year ago
Cyrilvallez fix typo
7c31343f
Cyrilvallez finally fix mask slicing
8004a706
Cyrilvallez Finally correctly slice in all cases!!
de59ddc8
Cyrilvallez add test for all attention functions
5667025e
Cyrilvallez small fix in tests
1db64c67
Cyrilvallez trick around dynamo tracing issue
f1d4868a
Cyrilvallez last update
ba84059b
Cyrilvallez more robust
0f0958fd
Cyrilvallez kwargs propagation
370326a5
Cyrilvallez make it explicit for checkpointing
213da5a0
Cyrilvallez apply modular
ee426a89
Cyrilvallez
Cyrilvallez
Cyrilvallez Cyrilvallez changed the title Fix FA2 for models with HybridCache Fix mask slicing for models with HybridCache 1 year ago
ArthurZucker
ArthurZucker commented on 2025-01-21
ArthurZucker
ArthurZucker approved these changes on 2025-01-28
Cyrilvallez Cyrilvallez merged 3f860dba into main 1 year ago
Cyrilvallez Cyrilvallez deleted the fix-fa2-hybrid branch 1 year ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone