PR #39121 [qwen2-vl] fix FA2 inference

[qwen2-vl] fix FA2 inference #39121

zucchini-nlp merged 9 commits into huggingface:main from zucchini-nlp:qwen2-fix

fix FA2

4301faab

zucchini-nlp requested a review from

Cyrilvallez 281 days ago

zucchini-nlp added for patch

vasqu commented on 2025-06-30

update is causal flag and remove mask for FA2

2cb7f219

vasqu commented on 2025-06-30

vasqu approved these changes on 2025-06-30

ArthurZucker commented on 2025-06-30

update for FA2 with varlen path

7823ffc5

Merge remote-tracking branch 'upstream/main' into qwen2-fix

5640fe1d

how the tests were passing with different devices?

ebb46c38

vasqu commented on 2025-06-30

add comment and ref to the PR

f2248b7b

vasqu approved these changes on 2025-07-01

move mask preparation to base pretrained model

c567fe65

seq len is the first dim, not second

3a3afe3e

ArthurZucker approved these changes on 2025-07-01

fix copies to fix GLM4V

0e1d6865

zucchini-nlp enabled auto-merge (squash) 280 days ago

zucchini-nlp merged 7a25f8df into main 280 days ago

Reviewers

ArthurZucker

vasqu

Cyrilvallez

Assignees

No one assigned

Labels

for patch

Milestone

No milestone