Is it not possible to set the rollout engine to multiple gpus?
Open
Nobody has claimed this yet.
question
- Dominant language
- Python
- Stars
- 10.8k
- Forks
- 989
- Avg merge
- 6h 29m
- Merged PRs (30d)
- 85
Description
When I set the 'tensor_parallel_size'>1 in the engine args the code crashes. I understand that the training itself with Unsloth only supports one gpu for now (though they have announced this is changing), but is it not possible to still perform the rollouts with the vllm engine across multiple gpus?
Contributor guide
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 by reproducing the crash with tensor_parallel_size greater than 1 in the rollout engine while training remains configured for one GPU. Trace how the engine arguments reach the vllm rollout setup and identify the limitation. Done means multi-GPU rollouts work without requiring multi-GPU Unsloth training, with coverage for the relevant configuration.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100