Maximilian Maag d85733f22b fix(installer): pytorch and ROCm versions are incompatible ...

Each version of torch is only available for specific versions of CUDA and ROCm.
The Invoke installer and dockerfile try to install torch 2.4.1 with ROCm 5.6
support, which does not exist. As a result, the installation falls back to the
default CUDA version so AMD GPUs aren't detected. This commits fixes that by
bumping the ROCm version to 6.1, as suggested by the PyTorch documentation. [1]

The specified CUDA version of 12.4 is still correct according to [1] so it does
need to be changed.

Closes #7006
Closes #7146

[1]: https://pytorch.org/get-started/previous-versions/#v241

2024-10-23 09:59:00 +11:00

..

__init__.py

doc updates; clean up install directory

2023-02-07 16:35:22 -05:00

installer.py

fix(installer): pytorch and ROCm versions are incompatible

2024-10-23 09:59:00 +11:00

main.py

tidy(installer): remove no-op --version CLI arg of installer

2024-03-26 14:48:29 +11:00

messages.py

feat(installer): add options to include or exclude xFormers based on the GPU model

2024-10-09 22:46:18 -04:00