lm-sys / lm-sys/FastChat

Slower throughput with openai_server

Open
#3,468 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
39.5k
Forks
4.8k
PR merge metrics
No merged PRs in 30d

Description

Hi,

I deployed the model backend using fastchat worker. I tested the throughput for llama3-8b model and it reached >2500 tokens/seconds on A100. However when I started using the same model through openai_server for openai compatible endpoint, the throughput reduced drastically to < 500 tokens/seconds. I understand that there would be some reduction but not this much. Has someone else faced the same problem. If yes, how did you optimize it.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the reported comparison for llama3-8b on an A100, testing the FastChat worker directly and through openai_server. Trace the openai_server request path and compare its throughput with the worker path; done means identifying the source of the drop below 500 tokens/second and documenting a verified optimization or limitation.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
api, backend, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.