Add ROCm 7.2 support and default to it#476
Conversation
|
✅ All contributors have signed the CLA. Thank you! This PR is ready to be merged. |
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: ASSERTIVE Plan: Pro Plus Run ID: 📒 Files selected for processing (6)
📝 WalkthroughWalkthroughThe default ROCm version used across comfy-cli was bumped from 6.3 to 7.2. A new ChangesROCm 7.2 default update
No sequence diagram needed here — just a version number getting its glow-up, from 6.3 to a snazzier 7.2! 🎉 Related Issues: Resolves ROCm outdated version concern ( Suggested Labels: enhancement, dependencies Suggested Reviewers: (based on ownership of comfy_cli/uv.py and comfy_cli/command/install.py) Poem 🚥 Pre-merge checks | ✅ 2✅ Passed checks (2 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
✨ Simplify code
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
I have read and agree to the Contributor License Agreement |
Closes #472.
comfy install --gpu amdcurrently defaults to ROCm 6.3, and theROCmVersionenum stops at 7.1, so there's no way to pull the current ROCm 7.2 PyTorch wheels. PyTorch now ships ROCm 7.2 wheels athttps://download.pytorch.org/whl/rocm7.2, so this adds7.2to the enum and makes it the default. Follows the same shape as #379 (which added the flag and bumped the default 6.1 → 6.3), including the test updates.Changes
constants.py: addv7_2 = "7.2"toROCmVersion.cmdline.py/command/install.py: bump defaultrocm_versionv6_3→v7_2.uv.py: bumprocmPytorchUrlandrocmTorchBackendrocm6.3→rocm7.2.tests/uv/test_uv.py:test_torch_backend_amdnow expects therocm7.2default backend.tests/comfy_cli/test_install_python_resolution.py: add the7.2 → whl/rocm7.2case to the parametrized URL test.Users can still pin an older build with
--rocm-version 7.1(or 6.x).Why bumping the default is safe for older cards
The ROCm 7.2 wheel is built for a wide arch list, so RDNA2/older cards stay covered, not just RDNA3/4: gfx900, gfx906, gfx908, gfx90a, gfx942, gfx950, gfx1030, gfx1100, gfx1101, gfx1102, gfx1150, gfx1151, gfx1200, gfx1201
Tested
Verified
torch 2.12.1+rocm7.2installs and initializes from the rocm7.2 index on two AMD systems:cuda.is_available()Unit tests pass locally:
pytest tests/uv/test_uv.py tests/comfy_cli/test_install_python_resolution.py→ 61 passed.