Skip to content

pytorch-rocm crashes on importing

Description:

Upon trying to do import torch, the interpreter crashes immediately:

kouta@arch-kouta ~ % python3
Python 3.12.3 (main, Apr 23 2024, 09:16:07) [GCC 13.2.1 20240417] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
python3: /usr/src/debug/hip-runtime-amd/clr-rocm-6.0.2/hipamd/src/hip_code_object.cpp:762: hip::FatBinaryInfo** hip::StatCO::addFatBinary(const void*, bool): Assertion `err == hipSuccess' failed.
[1]    25444 IOT instruction (core dumped)  python3

Other ROCm applications such as ollama and blender work correctly, the issue seems to be localized entirely within torch. It also started happening with the recent python3.12 update, however I don't have the necessary old packages to revert to the last 3.11 release and check if it works there.

Additional info:

  • package version(s):
    pacman -Q python-pytorch-rocm python hip-runtime-amd 
    python-pytorch-rocm 2.3.0-1
    python 3.12.3-1
    hip-runtime-amd 6.0.2-2
  • hardware: AMD Radeon RX 7600

Steps to reproduce:

  1. Open a python3 interpreter and run import torch
  2. Interpreter crashes instantly.
To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information