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.

Sign up for GitHub

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

Jump to bottom

Llama Vision PEFT #1937

Open
wants to merge 7 commits into
base: main
Choose a base branch
from
Open

Llama Vision PEFT #1937

wants to merge 7 commits into from

Conversation

pbontrager
Copy link
Contributor

@pbontrager pbontrager commented Oct 31, 2024 •
edited
Loading

Context

What is the purpose of this PR? Is it to

  • add a new feature
  • fix a bug
  • update tests and/or documentation
  • other (please add here)

Please link to any issues this PR addresses. #1779

Changelog

What are the changes made in this PR?

  • Added llama3_vision_tune_to_peft_adapter_weights
  • Integrate it into checkpoint

Test plan

Please make sure to do each of the following if applicable to your PR. If you're unsure about any one of these just ask and we will happily help. We also have a contributing page for some guidance on contributing.

  • run pre-commit hooks and linters (make sure you've first installed via pre-commit install)
  • add unit tests for any new functionality
  • update docstrings for any new or updated methods or classes
  • run unit tests via pytest tests
  • run recipe tests via pytest tests -m integration_test
  • manually run any new or modified recipes with sufficient proof of correctness
  • include relevant commands and any other artifacts in this summary (pastes of loss curves, eval results, etc.)
from peft import PeftModel
from transformers import MllamaForConditionalGeneration 

model_id = "meta-llama/Llama-3.2-11B-Vision-Instruct"
peft_model_dir = "/tmp/Llama-3.2-11B-Vision-Instruct/"

model = MllamaForConditionalGeneration.from_pretrained(model_id)
peft_model = PeftModel.from_pretrained(model, peft_model_dir)

@pytorch-bot PyTorch Bot
Copy link

pytorch-bot bot commented Oct 31, 2024 •
edited
Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchtune/1937

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

✅ No Failures

As of commit 8d960b5 with merge base e1caa9f (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 Oct 31, 2024
@pbontrager pbontrager marked this pull request as ready for review November 13, 2024 18:38
@pbontrager pbontrager changed the title [WIP] Llama Vision PEFT Llama Vision PEFT Nov 13, 2024
ebsmothers
ebsmothers reviewed Nov 13, 2024
View reviewed changes
torchtune/models/llama3_2_vision/_convert_weights.py Outdated
Comment on lines 511 to 512
# if key == "decoder.layers.3.layer.attn.q_proj.lora_a.weight":
# import pdb; pdb.set_trace()
Copy link
Contributor

Choose a reason for hiding this comment

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

remove

ebsmothers
ebsmothers reviewed Nov 13, 2024
View reviewed changes
torchtune/training/checkpointing/_checkpointer.py
@@ -635,6 +635,7 @@ def save_checkpoint(
)

if training.ADAPTER_KEY in state_dict:
# import pdb; pdb.set_trace()
Copy link
Contributor

Choose a reason for hiding this comment

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

remove

ebsmothers
ebsmothers approved these changes Nov 13, 2024
View reviewed changes
Copy link
Contributor

@ebsmothers ebsmothers left a comment

Choose a reason for hiding this comment

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

Run a forward parity check please. Only one comment, otherwise looks good

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@ebsmothers ebsmothers ebsmothers approved these changes

Assignees
No one assigned
Labels
CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@pbontrager @ebsmothers @facebook-github-bot