-
Notifications
You must be signed in to change notification settings - Fork 180
[Examples] Fix ignore layers for Qwen2.5-VL #1658
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
base: main
Are you sure you want to change the base?
Conversation
fixes Some layers that were to be ignored were not found in the model: {'re:visual.*'} as the vision module in Qwen2.5-vl is in model.visual
👋 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. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Summary of Changes
Hello @SorenDreano, 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 and resolves a warning encountered during the quantization process for the Qwen2.5-VL model example. The issue stemmed from an incorrect regular expression pattern in the module ignore list, which was attempting to exclude layers that did not exist under that specific path. By refining this configuration, the example now runs without the previous warning, ensuring a cleaner and more accurate setup for model quantization.
Highlights
- Configuration Fix: I've corrected the
ignore
list within theGPTQModifier
configuration for the Qwen2.5-VL example. Specifically, there:visual.*
pattern, which was causing a warning, has been removed as it was redundant and incorrect for the model's structure.
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 configure Gemini 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.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Code Review
The pull request correctly removes a redundant entry from the ignore
list in the Qwen2.5-VL example, which was causing a warning. I've suggested a small improvement to make the remaining regex pattern more precise to avoid potential issues with overly broad matching.
Hi @SorenDreano, Thank you for your contribution! However, in this case, these ignore names are kept in place in order to support older checkpoints created prior to huggingface/transformers#37033. Perhaps instead of removing them, we can add a comment in the example explaining why they exist. EDIT: See @brian-dellabetta's review |
@@ -73,7 +73,7 @@ def data_collator(batch): | |||
GPTQModifier( | |||
targets="Linear", | |||
scheme="W4A16", | |||
ignore=["lm_head", "re:visual.*", "re:model.visual.*"], | |||
ignore=["lm_head", "re:model.visual.*"], |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Since we want to catch both visual.
and model.visual.
, we can use an OR condition in our regex like
ignore=["lm_head", "re:model.visual.*"], | |
# NOTE: catch anything beginning with `model.visual.` or `visual.` | |
# to account for different behavior across transformers versions | |
ignore=["lm_head", "re:(model\.|)visual\..*"], |
This will prevent the warning from appearing
SUMMARY:
Oneshot() prints the following warning:
Some layers that were to be ignored were not found in the model: {'re:visual.*'} as the vision module in Qwen2.5-vl is in model.visual
TEST PLAN:
Removing re:visual.* from the list of modules to be ignored did not result in any warning or error