HiCache storage, style change and bug fix #8719
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @xiezhq-hermann, 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!
I've refactored the 'HiCacheStorage' implementation, specifically 'HiCacheFile', to improve how tensors are stored and retrieved. The primary goal was to streamline the I/O operations by moving away from 'torch.save' and 'torch.load' towards direct buffer manipulation, which should offer more control and potentially better performance for HiCache operations. This also includes a bug fix related to how 'target_location' is handled during tensor retrieval.
Highlights
- HiCache Storage Refinement: I've updated the 'HiCacheFile.get' method to consistently load tensor data directly into a provided 'target_location', removing the previous 'torch.load' fallback. This ensures more predictable memory handling.
- Optimized Tensor Saving: I've changed the 'HiCacheFile.set' method to save tensors directly to file using 'value.contiguous().view(dtype=torch.uint8).numpy().tofile(tensor_path)', replacing the less efficient 'torch.save'.
- Type Hint Improvements: I've refined the type hints for 'target_location' in 'get' and 'batch_get' methods to explicitly 'torch.Tensor' or 'List[torch.Tensor]', improving code clarity and type safety.
- Minor Code Cleanup: I've removed an outdated 'todo' comment regarding tensor object access for different TP ranks and updated '.storage()' to '.untyped_storage()' for better compatibility.
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 several changes to the HiCacheFile storage backend. It updates the get and set methods to use raw byte storage instead of torch.save/torch.load, which is a good performance improvement. It also updates method signatures for better type safety and replaces a deprecated PyTorch API call. My review has identified two critical issues. First, the set method contains invalid syntax that will cause a runtime error. Second, while the signature of batch_get was updated, its implementation was not, leading to an inconsistency that will also cause a runtime error under certain conditions. I have provided specific comments and a suggestion for one of the issues.
Motivation
Following #7704, #7211, #7280 and #8488.
Modifications
Accuracy Test
Benchmark & Profiling
Checklist