open-compass / open-compass/VLMEvalKit

Qwen2.5-VL-3B model evaluating BLINK, and it reported an OOM error:

Open
#1,416 0 comments 0 reactions 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

I am using the 910B NPU with the Qwen2.5-VL-3B model to evaluate BLINK, and it reported an error:
[2026-01-25 15:38:11] ERROR - RUN - run.py: main - 2090: Model Qwen2.5-VL-3B-Instruct x Dataset VStarBench combination failed: NPU out of memory. Tried to allocate 5.51 GiB (NPU 0; 60.96 GiB total capacity; 53.99 GiB already allocated; 53.99 GiB current active; 4.53 GiB free; 55.50 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.

How can I resolve this issue? Even when I use 4*910B, it doesn't seem to distribute the memory usage evenly.

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 run.py at main line 2090 and reproduce the Qwen2.5-VL-3B-Instruct with the VStarBench evaluation described in the report. Trace how multi-device execution assigns model and evaluation memory, then document a verified configuration or change that avoids the NPU out-of-memory failure and distributes usage as expected.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.