open-compass / open-compass/VLMEvalKit

测试速度问题

Open
#1,072 2 comments 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
4.4k
Forks
768
Avg merge
1d 10h
Merged PRs (30d)
17

Description

感谢您的工程。尝试使用VLMEvalKit库测试在垂直领域微调后的qwen2.5-vl系列模型,使用指令python run.py --data MMBench_DEV_EN --model Qwen2.5-VL-3B-Instruct --verbose,其中模型配置了本地的地址,每次单独测试1个未微调/微调的权重,发现推理速度特别慢,显示8张卡都有显存占用,但是sm绝大部分都是0,只有偶尔1个1的情况,显示时间估计要几十个小时。之后限定单张卡情况下,显示时间还缩短了1/3左右。

问题:
1.想问一下您这个可能是什么情况呢(之前在别的服务器上测试qwen2.5-vl-3b和使用molmo,速度都比较快,显示1个小时内)?
2.问一下您使用lmdeploy、vllm加速的时候,看到不同文档中有两种写法,可以选择启动服务或不启动,直接调用--use-vllm,这个一般建议使用哪一种呢?
3.问一下--api-nproc 是不是类似起到了多批次的效果?
4.llm那边使用opencompass测试多次重复实验的结果不一致,一般是怎么处理这个随机性呢?

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 with the run.py entry point and the documented --use-vllm and --api-nproc options, then compare the local Qwen2.5-VL evaluation path with the lmdeploy, vllm, and OpenCompass usage mentioned in the report. Done would require a reproducible explanation of the low GPU utilization and guidance on acceleration, parallelism, and repeatability.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
machine-learning, performance, testing-qa
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
18/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.