Skip to content

Add lora_path argument to bench_multiturn.py#10092

Merged
zhyncs merged 2 commits intosgl-project:mainfrom
Fridge003:lora_bench
Sep 6, 2025
Merged

Add lora_path argument to bench_multiturn.py#10092
zhyncs merged 2 commits intosgl-project:mainfrom
Fridge003:lora_bench

Conversation

@Fridge003
Copy link
Copy Markdown
Collaborator

Motivation

Used for benchmarking lora feature with radix cache
Usage:

python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --lora-paths lora1=algoprog/fact-generation-llama-3.1-8b-instruct-lora

python3 benchmark/hicache/bench_multiturn.py --model-path meta-llama/Llama-3.1-8B-Instruct  --lora-path lora1

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot left a 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 @Fridge003, 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 enhances the bench_multiturn.py benchmarking script by introducing support for LoRA (Low-Rank Adaptation) paths. This allows users to benchmark the performance of specific LoRA adaptors, particularly in scenarios involving radix cache, by integrating the LoRA path directly into the request generation process.

Highlights

  • New command-line argument for LoRA path: The bench_multiturn.py script now accepts a --lora-path argument, allowing users to specify a single LoRA adaptor for benchmarking.
  • LoRA path integration into request payload: The gen_payload function has been updated to accept a lora_path argument, which is then included in the request payload sent to the SGLang server.
  • Dynamic LoRA path usage in benchmarking: The lora_path provided via command-line is now dynamically passed to the gen_payload function during initial request generation and subsequent response handling within the benchmark process.
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. 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

  1. 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.

Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request adds a --lora-path argument to bench_multiturn.py to facilitate benchmarking with LoRA adapters. The changes correctly thread this new argument through to the payload generation. However, I've found a critical scope issue where the args object is accessed in a method where it's not available, which will cause a runtime error. My review includes a comment detailing this issue and how to fix it.

gen_payload(
self.client_records[client_id]["history"],
self.output_length,
args.lora_path,
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

critical

The args variable is not accessible within the response_handler method. It's a local variable from the main execution block and is passed to WorkloadGenerator.__init__, but it is not stored as an instance attribute. Accessing args.lora_path here will result in a NameError.

To fix this, you should store lora_path on the instance (self) within the __init__ method and use self.lora_path here.

In WorkloadGenerator.__init__:

# In WorkloadGenerator.__init__(self, args)
self.lora_path = args.lora_path

Then, update this call to gen_payload:

gen_payload(
    self.client_records[client_id]["history"],
    self.output_length,
    self.lora_path,
),

@Fridge003
Copy link
Copy Markdown
Collaborator Author

cc @xiezhq-hermann

@zhyncs zhyncs merged commit beac202 into sgl-project:main Sep 6, 2025
19 checks passed
@Fridge003 Fridge003 deleted the lora_bench branch September 8, 2025 05:01
MahmoudAshraf97 pushed a commit to MahmoudAshraf97/sglang that referenced this pull request Sep 8, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants