apache / apache/arrow

[Python][Docs] Improve the Python user guide on the CUDA integration (`pyarrow.cuda`)

Open
#41,666 0 comments 0 reactions 0 assignees View on GitHub
Component: Documentation Component: GPU Component: Python
Dominant language
C++
Stars
17.1k
Forks
4.3k
Avg merge
3d 13h
Merged PRs (30d)
88

Description

We have a page in the user guide about the CUDA integration with the `pyarrow.cuda` module: https://arrow.apache.org/docs/15.0/python/integration/cuda.html. But this page is quite brief and outdated, even for the current state of the CUDA functionality (for example it only briefly shows buffers, but doesn't mention anything about having an Array or RecordBatch on the CUDA device, or copying reading/writing directly from/to IPC, etc).

A list of ideas:

- Show reading/writing IPC
- Show how to copy full Array or RecordBatch to/from host (depending on improvements in https://github.com/apache/arrow/issues/41126)
- Expand section on interoperability with other tools (right now it only explains interop with numba, but could also add interop with eg pytorch or cupy, cudf, nanoarrow, etc)
- Add guide on installing pyarrow with CUDA enabled (from binaries, not to have to build yourself, e.g. this is possible through conda-forge)

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.