NVIDIA-NeMo / NVIDIA-NeMo/Curator
Batch size tuning for Hugging Face text classifiers
Open
@VibhuJawa is already working on this.
Since Dec 8, 2025.
enhancement
- Dominant language
- Python
- Stars
- 1.8k
- Forks
- 328
- Avg merge
- 4d 5h
- Merged PRs (30d)
- 30
Description
PR https://github.com/NVIDIA-NeMo/Curator/pull/753 migrates our multi-node, multi-GPU classifiers from CrossFit (Dask) to Ray. Although it does not use CrossFit anywhere, it takes from core CrossFit ideas, specifically sorting and trimming tokens for faster model inference.
The main CrossFit feature missing from the Ray PR is automatic batch size tuning. For now, the user can specify a batch size for the model inference step, but we should add more support to generate a reasonable batch size based on available hardware, model size, etc.
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.
Assessment
This issue has not been assessed yet.