TheRock: Developer builds from PRs and PyTorch packages

gistfile0.md

TheRock: Developer Builds from PRs and PyTorch Packages

Claude Code Prompt: Read the docs and workflows in therock using subagents and tell me how to make developer builds from a PR and get packages w/ pytorch.

Triggering PR Builds

Label-based (simplest)

Add labels to your PR:

• gfx120X, gfx94X, etc. - build for specific GPU family
• test:rocthrust, test:hipblaslt - run specific tests
• skip-ci - skip builds

Manual dispatch

Actions → CI workflow → Run workflow → specify linux_amdgpu_families

Getting Artifacts from a PR Build

Use build_tools/install_rocm_from_artifacts.py:

python build_tools/install_rocm_from_artifacts.py \
    --run-id [RUN_ID_FROM_ACTIONS_URL] \
    --amdgpu-family gfx110X-all \
    --blas --rccl --tests

Or use nightly tarballs directly:

python build_tools/install_rocm_from_artifacts.py \
    --release 7.11.0a20251124 \
    --amdgpu-family gfx110X-all

PyTorch from Dev Builds

Two-step process:

1. Copy ROCm packages to dev bucket via copy_release.yml workflow
2. Trigger PyTorch build via build_portable_linux_pytorch_wheels.yml:
   • Set release_type: dev (important - not nightly/prerelease)
   • Specify rocm_version matching what you copied
   • Wheels land at https://rocm.devreleases.amd.com/v2/{GPU_FAMILY}/

Key Docs

• docs/development/installing_artifacts.md
• docs/development/ci_behavior_manipulation.md
• docs/development/github_actions_debugging.md