Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Idefics2 #1270

Merged
merged 14 commits into from
Nov 7, 2024
Merged

Idefics2 #1270

merged 14 commits into from
Nov 7, 2024

Conversation

sywangyi
Copy link
Collaborator

What does this PR do?

Fixes # (issue)

Before submitting

  • This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
  • Did you make sure to update the documentation with your changes?
  • Did you write any new necessary tests?

Signed-off-by: Wang, Yi A <[email protected]>
Signed-off-by: Wang, Yi A <[email protected]>
Signed-off-by: Wang, Yi A <[email protected]>
@HuggingFaceDocBuilderDev

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

@sywangyi
Copy link
Collaborator Author

sywangyi commented Aug 19, 2024

throught(tokens/s) A100 Gaudi2
BF16 11.7 53.6
FP32 15.3 29.53

cmd:
python3 run_pipeline.py --model_name_or_path HuggingFaceM4/idefics2-8b --use_hpu_graphs --bf16

Copy link
Collaborator

@ssarkar2 ssarkar2 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good otherwise, have a question regd unfold

Signed-off-by: Wang, Yi A <[email protected]>
Copy link

github-actions bot commented Sep 6, 2024

The code quality check failed, please run make style.

Signed-off-by: Wang, Yi A <[email protected]>
move vision model to prepare input(image size may change if the input is dataset, can not use hpu graph to optimize it)
add ci test

Signed-off-by: Wang, Yi A <[email protected]>
Signed-off-by: Wang, Yi A <[email protected]>
Signed-off-by: Wang, Yi A <[email protected]>
@sywangyi
Copy link
Collaborator Author

Looks good otherwise, have a question regd unfold

Hi, I have replace unfold with conv2D. cpu fallback is not needed. finetune got speedup @ssarkar2

@sywangyi
Copy link
Collaborator Author

has found the recent PR which upgrade test_image_to_text_example.py. add "use_kv_cache", whose default value is false if not indicated explicitly in command line. which will lead image-to-text ci test to fail. fix it.

@sywangyi
Copy link
Collaborator Author

also found the similar PR which mean to fix eos issue if batch>1, please refer #1287,
the root cause of it is the needs_tensor_output is incorrect set

@libinta libinta added run-test Run CI for PRs from external contributors and removed review wip labels Oct 15, 2024
@sywangyi
Copy link
Collaborator Author

sywangyi commented Nov 7, 2024

@regisss could you review and merge it?

@regisss regisss merged commit 7ac1db1 into main Nov 7, 2024
7 checks passed
@regisss regisss deleted the idefics2 branch November 7, 2024 20:26
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
run-test Run CI for PRs from external contributors
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants