random #415
random #415
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 @chengtao-lv, 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 integrates a new token reduction method, RandomPrune, into the llmc library. This method focuses on randomly pruning image tokens to potentially optimize model performance or reduce computational overhead. The implementation ensures compatibility with different Llava model architectures by correctly identifying and adjusting relevant input tensors after pruning.
Highlights
- New Token Reduction Strategy: Introduces
RandomPrune, a new token reduction module that implements a random pruning strategy specifically for image tokens within a sequence. - Dynamic Image Token Pruning: The
RandomPruneclass calculates a number of image tokens to keep based on a specified rate, then randomly selects these tokens. It meticulously updates thehidden_states,position_ids,attention_mask,cache_position, andposition_embeddingsto reflect the pruned sequence, ensuring model integrity. - Llava Model Integration: Includes specific forward pre-hooks for
LlavaHfandLlavamodels to correctly identify the starting index and length of image tokens, enabling accurate application of the random pruning logic.
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.
Code Review
This pull request introduces a new token reduction method, RandomPrune, which randomly prunes vision tokens for Llava-like models. The implementation uses PyTorch hooks to modify hidden states and related tensors during the forward pass. My review identified a couple of critical correctness issues, including a potential IndexError and a typo in a tensor operation, which could lead to runtime crashes or incorrect model behavior.
| pruning_paras['image_token_start_index'] = torch.where(token_indices)[ | ||
| 0 | ||
| ][0].item() |
There was a problem hiding this comment.
This line can raise an IndexError if torch.where(token_indices)[0] returns an empty tensor, which happens when no IMAGE_TOKEN_INDEX is found in the input. This would crash the forward pass. You should add a check to ensure that at least one image token index is found before attempting to access it.
| image_token_idxs = ( | ||
| input_ids[0] == pruning_paras['vision_token_index'] | ||
| ).nonzero(as_tuple=True)[0] | ||
| pruning_paras['image_token_start_index'] = image_token_idxs[0].item() |
There was a problem hiding this comment.
This line can raise an IndexError if input_ids[0] == pruning_paras['vision_token_index'] returns an empty tensor, which happens when no vision_token_index is found in the input. This would crash the forward pass. You should add a check to ensure that at least one image token index is found before attempting to access it.
| new_pe0 = position_embeddings[0][:, keep_indexs, :].clone() | ||
| new_pe1 = position_embeddings[1][:, keep_indexs, :].clone() | ||
| position_embeddings[0].resize_as_(new_pe0).copy_(new_pe0) | ||
| position_embeddings[1].resize_as_(new_pe0).copy_(new_pe1) |
There was a problem hiding this comment.
There appears to be a typo here. position_embeddings[1] is being resized using the shape of new_pe0 instead of new_pe1. While their shapes might be identical in the current implementation, this is semantically incorrect and could lead to hard-to-debug errors if the shapes ever differ.
position_embeddings[1].resize_as_(new_pe1).copy_(new_pe1)
No description provided.