deepspeedai / deepspeedai/DeepSpeed
[BUG]CUDA error in pipeline parallel
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
Describe the bug
When I trained the model with two nodes for pipeline parallel tasks, each node has eight graphics cards. So the incoming LOCAL_RANK of Node One is 8-17, and the line 201 of deepspeed/runtime/pipe/module. py
self. to (get_accelerator().device_name (self.local_rank)).
Here local_rank does not match the graphics card numbers 0-7, so CUDA error is raised
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 at deepspeed/runtime/pipe/module.py line 201 and inspect how LOCAL_RANK is passed to the device selection for pipeline parallel training. Reproduce the setup with two nodes and eight graphics cards per node; done means the second node's ranks map to valid local devices without a CUDA error.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100