NVIDIA / NVIDIA/TransformerEngine
Improve comm_gemm_overlap documentation and HuggingFace integration guidance
Open
Nobody has claimed this yet.
documentation
- Dominant language
- Python
- Stars
- 3.5k
- Forks
- 831
- Avg merge
- 3d 11h
- Merged PRs (30d)
- 65
Description
The comm_gemm_overlap example needs clearer documentation on:
- How to enable GEMM+communication overlap - The process (initialize userbuffers, set layer flags, cleanup) should be explicitly documented with minimal code snippets
- Layer-specific parameters - Clarify the difference between ub_tp_comm_overlap vs individual flags
- Hugging Face integration - Provide guidance on combining with HF model replacement patterns
Current state:
examples/pytorch/comm_gemm_overlap/README.md covers requirements and run commands but lacks setup guidance
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 with examples/pytorch/comm_gemm_overlap/README.md and review the existing comm_gemm_overlap example context. Document initialization, layer flags, cleanup, the difference between ub_tp_comm_overlap and individual flags, and Hugging Face model replacement guidance with minimal snippets. Done means the README explains setup and integration in addition to its existing requirements and run commands.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- huggingface, python, pytorch
- Domain
- documentation, machine-learning
- Issue type
- Documentation
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 52/100