Skip to content

fix: bump launcher ROCm torch index from rocm6.3 to rocm7.1#9337

Open
chejh-amd wants to merge 1 commit into
invoke-ai:mainfrom
chejh-amd:main
Open

fix: bump launcher ROCm torch index from rocm6.3 to rocm7.1#9337
chejh-amd wants to merge 1 commit into
invoke-ai:mainfrom
chejh-amd:main

Conversation

@chejh-amd

Copy link
Copy Markdown

pins.json still points the linux rocm torch index at https://download.pytorch.org/whl/rocm6.3, while pyproject.toml has already moved the torch-rocm index to rocm7.1. So the launcher installs rocm6.3 wheels even though the source/pip install uses rocm7.1. This is the mismatch reported in #9328 (the launcher runs --index=https://download.pytorch.org/whl/rocm6.3).

This is especially broken on RDNA4: rocm6.3 wheels predate gfx1201 support (RDNA4 needs ROCm 7.0+), so they can't work on cards like the RX 9070 XT / R9700 at all.

This bumps the rocm entry in pins.json to rocm7.1 so the launcher matches pyproject.toml.

Tested on AMD Radeon AI PRO R9700 (gfx1201, RDNA4), Linux:

  • pip install torch torchvision --index-url https://download.pytorch.org/whl/rocm7.1torch 2.12.1+rocm7.1, torch.cuda.is_available() = True (AMD Radeon AI PRO R9700)

Fixes #9328.

@github-actions github-actions Bot added the Root label Jul 6, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[bug]: Release 6.13.5rc1 not installing ROCm 7.1 as expected

1 participant