You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
On image rocm/pytorch:rocm6.1_ubuntu22.04_py3.10_pytorch_2.4 (and probably all the others) pip check returns dependencies errors. This leads to pip reinstalling torch without ROCm support when any pip package which lists it as a dependency is installed. In particular, python-triton-rocm is not installed, and the wrong version of numpy is installed. See also pypa/pip#12902.
pip check output:
$ docker run --rm -it rocm/pytorch:rocm6.1_ubuntu22.04_py3.10_pytorch_2.4 bash
root@2857e8f82486:/var/lib/jenkins# pip check
torch 2.4.0.dev20240520+rocm6.0 requires pytorch-triton-rocm, which is not installed.
numba 0.55.2 has requirement numpy<1.23,>=1.18, but you have numpy 1.26.4.
scipy 1.8.1 has requirement numpy<1.25.0,>=1.17.3, but you have numpy 1.26.4.
Versions
No response
The text was updated successfully, but these errors were encountered:
🐛 Describe the bug
On image
rocm/pytorch:rocm6.1_ubuntu22.04_py3.10_pytorch_2.4
(and probably all the others)pip check
returns dependencies errors. This leads to pip reinstallingtorch
without ROCm support when any pip package which lists it as a dependency is installed. In particular,python-triton-rocm
is not installed, and the wrong version ofnumpy
is installed. See also pypa/pip#12902.pip check
output:Versions
No response
The text was updated successfully, but these errors were encountered: