Changes/PyTorch2.4

PyTorch 2.4

This is a proposed Change for Fedora Linux.
This document represents a proposed Change. As part of the Changes process, proposals are publicly announced in order to receive community feedback. This proposal will only be implemented if approved by the Fedora Engineering Steering Committee.

Summary

PyTorch is a popular python library for deep learning using GPUs and CPUs. It is frequently associated with various AI implementations and other neural network based projects. This change will update PyTorch to the latest upstream version (2.4).

Owner

• Name: Tom Rix
• Email: trix@redhat.com

Current status

• Targeted release: Fedora Linux 41
• Last updated: 2024-07-17
• [Announced]
• [<will be assigned by the Wrangler> Discussion thread]
• FESCo issue: <will be assigned by the Wrangler>
• Tracker bug: <will be assigned by the Wrangler>
• Release notes tracker: <will be assigned by the Wrangler>

Detailed Description

Update the base Fedora PyTorch from 2.2 to the latest upstream 2.4. The big change for 2.4 is the integration with the ROCm stack to provide open accelerated AI on AMD GPU's.

Benefit to Fedora

Fedora AI is keeping pace with the fast moving AI by updating PyTorch to match what is available *right now* in the upsteam with features needed to do real AI development for open source.

Scope

• Proposal owners: Update base to 2.4 when upstream releases.

• Other developers:

• Release engineering: N/A (not needed for this Change)

• Policies and guidelines: N/A (not needed for this Change)

• Trademark approval: N/A (not needed for this Change)

• Alignment with the Fedora Strategy: "Innovation & Leadership in Technology: We integrate programming language stack ecosystems" by including the latest release of PyTorch, Fedora will have the latest release of a popular software stack in AI/ML.

Upgrade/compatibility impact

There should be no backwards incompatible changes with the 2.4 release.

Early Testing (Optional)

Do you require 'QA Blueprint' support? N

How To Test

Testing can be done with CPU only but that may lead to very long test times. Testing with ROCm compatible hardware is recommended.

Running through some of the PyTorch Examples is a good way for basic testing of PyTorch. Running through the tests contained in other torch-related packages (e.g python-torchvision) could also be helpful.

User Experience

this section needs review to make sure there aren't changes which need to be outlined

Users will be able to continue to use PyTorch installed from the Fedora repos - the upgrades to llvm require a newer ROCm and that newer ROCm is only supported by the latest PyTorch release.

Other than being able to continue to use PyTorch from the Fedora repos, the user-noticeable changes will be minimal.

Dependencies

There are no RPMs affected by this change that are not included in the change.

Contingency Plan

Fedora currently ships with the latest public 2.3 release with all of the accelerated features.

• Contingency mechanism: (What to do? Who will do it?) N/A (not a System Wide Change)
• Contingency deadline: N/A (not a System Wide Change)
• Blocks release? No, not a System Wide Change.

Documentation

N/A (not a System Wide Change)