NVIDIA-NeMo / NVIDIA-NeMo/Curator
Cloud Storage Compatibility (fsspec) Throughout
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 1.8k
- Forks
- 328
- Avg merge
- 4d 5h
- Merged PRs (30d)
- 30
Description
Replace all os.path, pathlib.Path, glob, and shutil I/O in processing stages with fsspec-backed equivalents so S3, GCS, and Azure paths work identically to local paths.
Why: All production data lives in object storage. Today, Curator's download stages support cloud I/O but the processing stages use local-only primitives, breaking on s3:// or gs:// URIs. This forces users to sync data locally before processing — untenable at large scale.
Definition of Done:
- All file reads/writes in stages/ use fsspec.open() or equivalent
- s3://, gs://, az://, hf://, and file:// paths work in all stages without code changes
- FilePartitioningStage resolves cloud paths correctly (issue #1568)
- CI tests added for S3-compatible storage (e.g., MinIO in CI)
- No dependency on os.path, pathlib.Path, glob, or shutil for URI handling in stage code (enforced by linter rule per issue #984)
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 inventorying URI handling throughout stages/ and read FilePartitioningStage alongside issue #1568. Then inspect the CI setup for adding MinIO coverage. Done means all listed URI schemes work across stages, local-only path primitives are absent from stage URI handling, and the linter requirement is enforced.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- aws, azure, gcp, python
- Domain
- cloud, data-engineering
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100