Setup LM Studio to work with ROCM on RX6600 (GFX1032) on Windows
- Download and install HIP SDK 5.7.1 (direct link)
- Download amd-rocm-avx2-1.1.5
- Extract all files and put on
C:\Users\%USERNAME%\.cache\lm-studio\extensions\backends\amd-rocm-avx2-1.1.5
- Open backend-manifest.json with any text editor and add “gfx1032” in gpu.targets
- Download rocm-vendor-v2
- Extract all files to
C:\Users\%USERNAME%\.cache\lm-studio\extensions\backends\vendor\win-llama-rocm-vendor-v2
- Delete all files inside
rocblas\library
- Download Optimised_ROCmLibs_gfx1032
- Extract all files to
rocblas\library
- Download ollama-windows-amd64.7z (v0.3.6)
- Open
ollama-windows-amd64.7z\ollama_runners\rocm_v5.7
.
- Select the files
ggml.dll
and llama.dll
.
- Extract and replace these files in
C:\Users\%USERNAME%\.cache\lm-studio\extensions\backends\amd-rocm-avx2-1.1.5
.
- Open LM Studio
- Set LM Studio to Developer
- Go to Developer > LM Runtimes > Configure Runtimes
- Select llama.cpp-win-x86_64-amd-rocm-avx2
Now try to load some model and see if it will work