open-compass / open-compass/VLMEvalKit
Qwen2.5-VL-3B model evaluating BLINK, and it reported an OOM error:
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
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- 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