[qgallouedec]

Follow-up to #5918, which runs the test suite on AMD MI300X GPUs for every pull request.

Adds a "Training on AMD GPUs" page next to the Trainium one, and a line in the README and the docs index. The page covers what differs from CUDA: installing PyTorch's ROCm build, the AITER attention kernel in place of flash-attn2 (no ROCm build), which drops Gemma 2's attention soft-capping (huggingface/transformers#49810), and vLLM, which needs its own ROCm build and isn't in CI yet.

• On MI300X the suite gives the same results as the CUDA job: 2,834 passed and 213 skipped, and the 61 distributed tests pass.

---

Note

Low Risk
Documentation-only changes with no runtime, security, or training logic impact.

Overview
Adds first-class documentation for running TRL on AMD Instinct GPUs, mirroring the existing AWS Trainium how-to.

A new amd_gpus guide covers ROCm PyTorch install, using kernels-community/aiter-flash-attn instead of FlashAttention (and the Gemma 2 soft-capping caveat), and vLLM’s ROCm build for GRPO/RLOO. The docs toctree, README, and docs index “What’s New” sections link to that page and announce AMD support—no trainer or library code changes.

^{Reviewed by Cursor Bugbot for commit 148c39c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-20T22:22:32.381524Z	3656ffb	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3656ffb7c9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[bot-ci-comment]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[qgallouedec]

@kashif I'm merging this one so that I can advertise on the official support tomorrow