Support multiple tensor backends via NEP-47 (Python array API standard)
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 2.8k
- Forks
- 315
- Avg merge
- 3d 9h
- Merged PRs (30d)
- 27
Description
The recently proposed NEP-47 attempts to unify the APIs of various tensor frameworks (NumPy, Tensorflow, PyTorch, Dask, JAX, CuPy, MXNet, etc.), via the Python array API standard.
It is a much more compact version of the original NumPy APIs, removing unnecessary functions that are not friendly to heterogenous hardware like GPUs.
Since NumPyro is using JAX as backend, whose APIs closely match NumPy, it should be quite doable to adopt NEP-47 for multi-backend support?
Related discussion:
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 reading the linked NEP-47 and Python array API standard, then review the related array-api and PyMC discussions. Inspect how NumPyro currently relies on JAX as its backend. Done would require an agreed implementation of multi-backend support that follows the standard across the relevant tensor operations.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- numpy, python
- Domain
- backend, machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100