Nightly #382
Nightly #382
Conversation
* Handle import path change for vllm lora * Better handle sleep and wakeup * Revert "Better handle sleep and wakeup" This reverts commit 00a8d68.
Summary of ChangesHello @danielhanchen, 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 introduces crucial compatibility and stability improvements. It addresses a compilation issue with 8-bit layers in Highlights
Using Gemini Code AssistThe 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 by creating a comment using either
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 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. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. 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
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request includes two main changes. First, it adds a fix for compiling 8-bit LoRA layers by disabling torch.compile on bitsandbytes operations to prevent dimension errors. This is a good correctness fix. Second, it updates imports in the vllm_lora_worker_manager to handle a module reorganization in newer versions of vllm, ensuring continued compatibility. Overall, the changes are positive. I've provided one suggestion to use a public API for consistency in the compiler patch.
| extra_prepend = ( | ||
| "\nimport torch._dynamo\n" | ||
| "@torch._dynamo.disable\n" | ||
| "def _call_8bit_base_layer(base_layer, x, *args, **kwargs):\n" | ||
| " return base_layer(x, *args, **kwargs)\n" | ||
| ) |
There was a problem hiding this comment.
For consistency with other parts of the codebase and to prefer public APIs over internal ones, it would be better to use torch.compiler.disable instead of torch._dynamo.disable. torch.compiler.disable is the recommended public API for this functionality.
This change also requires updating the import from torch._dynamo to torch.compiler.
| extra_prepend = ( | |
| "\nimport torch._dynamo\n" | |
| "@torch._dynamo.disable\n" | |
| "def _call_8bit_base_layer(base_layer, x, *args, **kwargs):\n" | |
| " return base_layer(x, *args, **kwargs)\n" | |
| ) | |
| extra_prepend = ( | |
| "\nimport torch.compiler\n" | |
| "@torch.compiler.disable\n" | |
| "def _call_8bit_base_layer(base_layer, x, *args, **kwargs):\n" | |
| " return base_layer(x, *args, **kwargs)\n" | |
| ) |
No description provided.