OpenPipe / OpenPipe/ART

Cuda out of Memory

Open
#452 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

question
Dominant language
Python
Stars
10.8k
Forks
989
Avg merge
6h 29m
Merged PRs (30d)
85

Description

Hi guys,
I am trying to run Qwen 3B Instruct model on a GPU with 24 GB VRAM, but when the VLLM is creating CUDA graphs, it goes out of memory. It seems like we can set the gpu_memory_utilization config of VLLM to be around 0.7 to free up the GPU memory. Is there a way to pass this flag during backend initialization? Another interesting issue is that this happens when I run it on Databricks with a GPU of 24 GB VRAM, but when I run it on a local machine with RTX 3090, it runs fine. Not sure what the cause is. Thank you.

Contributor guide

Open the contributing guide

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 locating the backend initialization that creates the vLLM engine and check how gpu_memory_utilization can be passed. Compare the Qwen 3B Instruct run on Databricks with a 24 GB GPU against the RTX 3090 case. Done means the configuration can be supplied during initialization and the reported CUDA graph out-of-memory failure is resolved or clearly explained.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
ai, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.