deepspeedai / deepspeedai/DeepSpeed
Inference of model-parallel model on a single GPU
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
Hello. Is there any recommended way to save trained model-parallel model to a single checkpoint, that could be loaded to the single gpu for inference? Given that total amount of weights fits in single GPU memory using fp16.
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 reviewing DeepSpeed's model-parallel checkpointing and single-GPU inference entry points and documentation. Establish whether a supported conversion path exists for combining a model-parallel checkpoint into one checkpoint, and define successful loading and inference on a single GPU as the completion criteria.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- distributed-systems, machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 15/100