NOTE: that is kind of working, it will load model to GTT, but count available memory using VRAM
- Stop service:
sudo systemctl stop ollama - Update
/etc/systemd/system/ollama.serviceand add under[Service]Environment="HSA_OVERRIDE_GFX_VERSION=11.0.0"Environment="HCC_AMDGPU_TARGETS=gfx1100"
- Reload:
sudo systemctl daemon-reload - Start service:
sudo systemctl start ollama