Ollama fails to run on ROCm 6.2.2

Description:

The build of ollama-rocm in testing fails to run on an RX 7700 XT card, unlike upstream's binaries. It either results in a CUDA error, or a more verbose CUDA error.

Additional info:

Steps to reproduce:

  1. Install ollama-rocm and dependencies from testing
  2. Start ollama.service
  3. Run ollama run llama2
Edited by Christopher Snowhill