Skip to content

Unified Android aar support for llava and llama models #5086

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

Merged
merged 1 commit into from
Sep 5, 2024

Conversation

Riandy
Copy link
Contributor

@Riandy Riandy commented Sep 4, 2024

Summary:

  • Previously, we need two separate aar for vision and text models. Since ET core runner side has a combined aar built, I am making changes on the Android app side to support this behavior.
  • Introducing ModelUtils class so we can get the correct model category to be passed on to generate()
  • Seq_len is now an exposed parameters, defaulting to 128. For llava models, 128 is not enough, hence we are changing it to 768 when calling generate()
  • Minor bug fix on ETImage logic.

Reviewed By: cmodi-meta, kirklandsign

Differential Revision: D61406255

Copy link

pytorch-bot bot commented Sep 4, 2024

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/5086

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 45636f0 with merge base f326ee1 (image):
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

@facebook-github-bot facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Sep 4, 2024
@facebook-github-bot
Copy link
Contributor

This pull request was exported from Phabricator. Differential Revision: D61406255

Summary:
Pull Request resolved: pytorch#5086

- Previously, we need two separate aar for vision and text models. Since ET core runner side has a combined aar built, I am making changes on the Android app side to support this behavior.
- Introducing ModelUtils class so we can get the correct model category to be passed on to generate()
- Seq_len is now an exposed parameters, defaulting to 128. For llava models, 128 is not enough, hence we are changing it to 768 when calling generate()
- Minor bug fix on ETImage logic.

Reviewed By: cmodi-meta, kirklandsign

Differential Revision: D61406255
@facebook-github-bot
Copy link
Contributor

This pull request was exported from Phabricator. Differential Revision: D61406255

@facebook-github-bot facebook-github-bot merged commit cd1c833 into pytorch:main Sep 5, 2024
42 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. fb-exported
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants