PyTorch updates nvidia-cusparselt-cu13 to 0.8.1.post1

ResearchPublished: Source: PyTorch

#NVIDIA

Key point

An issue occurred where the aarch64 version of nvidia-cusparselt-cu13 0.8.1 had a platform tag in its metadata that did not comply with PEP 425. It has been updated to the corrected 0.8.1.post1.

It was reported that the aarch64 version of nvidia-cusparselt-cu13 0.8.1 had an issue where the tag 'py3-none-manylinux2014_sbsa' in .dist-info/WHEEL did not comply with PEP 425. This caused pip to install the wheel by its filename, but pip check to reject it, reporting it was not supported on the platform.

In response, corrected wheels were republished as 0.8.1.post1 with the tag 'py3-none-manylinux2014_aarch64'. Because '==0.8.1' does not match a post-release under PEP 440, PyTorch has bumped the five cu13 pins to 0.8.1.post1.

Version 0.8.1.post1 publishes three platform wheels (manylinux2014_aarch64, manylinux2014_x86_64, and win_amd64), moving the Linux and Windows pins together. The payload is byte-identical to 0.8.1, and the sha256 of nvidia/cusparselt/lib/libcusparseLt.so.0 is the same in both wheels. Only the .dist-info differs.

This item is a summary. The full, checkable version is at the original source:

Read the original at PyTorch ↗