[Example] Fix Qwen VL ignore list #1545
Conversation
|
👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review. Note: This is required to complete the testing suite, please only add the label once the PR is code complete and local testing has been performed. |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @arunmadhusud, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
This pull request addresses a change in the Hugging Face Transformers library (v4.52+) regarding the naming of weights for Qwen2-VL and Qwen2.5-VL models. To maintain compatibility and correctly prevent the vision model weights from being quantized, I've updated the quantization configuration in the relevant example scripts.
Highlights
- Quantization Ignore List: Updated the quantization ignore list for Qwen2-VL and Qwen2.5-VL examples to include the new weight naming convention introduced in Transformers v4.52.
- Vision Model Exclusion: Ensured that the vision components of the Qwen VL models are explicitly excluded from quantization by adding the 're:model.visual.*' pattern to the ignore list.
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist is currently in preview and may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments to provide feedback.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configureGemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
There was a problem hiding this comment.
Code Review
This pull request updates the ignore list for quantization in the Qwen2-VL and Qwen2.5-VL example scripts. The change adds the pattern "re:model.visual.*" to the list. This addition is necessary to correctly ignore vision model weights due to a naming convention change in transformers library version 4.52 and later, as detailed in the PR description.
The modification is applied consistently in both affected example files (qwen2_vl_example.py and qwen_2_5_vl_example.py). The change appears correct and directly addresses the stated problem, ensuring that the vision components of these models are not inadvertently quantized when using newer versions of the transformers library. The test plan of running the example scripts is appropriate for verifying this type of change. No issues of medium or higher severity were identified.
|
Thank you @arunmadhusud for the contribution! We were running into this issue and needed to debug this. I will validate locally and try to get this in if it's all good |
|
Resolves neuralmagic/compressed-tensors#353 |
There was a problem hiding this comment.
Tested locally. I was able to reproduce the error on latest transformers, and confirm that the example file runs successfully on this branch.
This resolves neuralmagic/compressed-tensors#353
SUMMARY: - Update ignore list as per #1545 - Note: We don't use an ignore list for for the Qwen2.5-VL FP8_Dynamic test?
**SUMMARY:** - Transformers v4.52 will map weights name for Qwen2-VL/Qwen2.5-VL: https://github.com/huggingface/transformers/blob/de4cf5a38e9678b9e465867a8a6b88ea727bea52/src/transformers/models/qwen2_vl/modeling_qwen2_vl.py#L1359-L1362 - So the qwen2-vl/qwen2.5-vl checkpoints saved after transformers v4.52 will have different weights name. - Update the ignore list to prevent the vision model from being quantized **CHANGES** - Add "model.visual." to ignore list **TEST PLAN:** Run [qwen2_vl_example.py](https://github.com/vllm-project/llm-compressor/blob/main/examples/multimodal_vision/qwen2_vl_example.py) Run [qwen_2_5_vl_example.py](https://github.com/vllm-project/llm-compressor/blob/main/examples/multimodal_vision/qwen_2_5_vl_example.py)
SUMMARY: - Update ignore list as per vllm-project#1545 - Note: We don't use an ignore list for for the Qwen2.5-VL FP8_Dynamic test?
SUMMARY:
CHANGES
TEST PLAN:
Run qwen2_vl_example.py
Run qwen_2_5_vl_example.py