update multi turn question eval #422
Conversation
![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 @helloyongyang, 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 significantly enhances the framework's capability to handle multi-turn question answering for vision-language models by ensuring that token reduction strategies maintain consistency across conversational turns. It introduces a new configuration for multi-turn evaluation and modifies existing token reduction algorithms to cache and reuse pruning decisions from the first turn, leading to more stable and coherent interactions.
Highlights
- Multi-Turn Evaluation Configuration: Added a new YAML configuration file (
sparsevlm_multi_turn.yml) specifically designed for evaluating SparseVLM in multi-turn question answering scenarios, including settings for Llava models, custom generation, and token reduction parameters. - Consistent Token Reduction Across Turns: Implemented logic across various token reduction methods (FastV, Random, SparseVLM, VisionZIP) to ensure consistency in token selection and masking for multi-turn conversations. This is achieved by introducing a
first_turn_questionflag and caching/reusing the token selection/masking policies from the initial turn for subsequent turns. - Model State Management: Introduced a
first_turn_questionattribute in the Llava model class (llmc/models/llava.py) to control the conditional logic for token reduction, ensuring that token pruning decisions are made once per conversation and then reused.
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 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 or fill out our survey 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.
Code Review
This pull request introduces support for multi-turn question evaluation, primarily for the LLaVA model. The changes involve adding a first_turn_question flag to differentiate between the first and subsequent turns in a conversation. This allows for caching and reusing computed values like attention masks and token indices in various token reduction methods (FastV, RandomPrune, SparseVLM, VisionZip), improving efficiency for multi-turn scenarios. The implementation appears correct and consistent across the modified files. My review includes one suggestion to improve configuration portability by replacing a hardcoded path with a placeholder.
| name: custom_gen | ||
| type: just_infer | ||
| download: False | ||
| path: /data/nvme1/yongyang/projects/llmc_plus/general_custom_data |
There was a problem hiding this comment.
The path for the evaluation data is hardcoded to a user-specific absolute path. This reduces the portability of the configuration file and makes it difficult for other developers to run this configuration without modification.
For better reusability, please consider using a placeholder, similar to how model.path and save.save_path are configured in this file.
path: /path/to/general_custom_data
No description provided.