Skip to content

Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

remove `GGML_CUDA_FORCE_MMQ=on` from build (#5588)

Jeffrey Morgan4e262eb2a8aa
diff --git a/llm/generate/gen_linux.sh b/llm/generate/gen_linux.sh
index 304eadbd9..5589f1ead 100755
--- a/llm/generate/gen_linux.sh
+++ b/llm/generate/gen_linux.sh
@@ -178,7 +178,7 @@ if [ -z "${OLLAMA_SKIP_CUDA_GENERATE}" -a -d "${CUDA_LIB_DIR}" ]; then
CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} ${OLLAMA_CUSTOM_CUDA_DEFS}"
echo "Building custom CUDA GPU"
else
- CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_FLAGS=-t8 -DGGML_CUDA_FORCE_MMQ=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} -DCMAKE_LIBRARY_PATH=/usr/local/cuda/compat"
+ CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_FLAGS=-t8 -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} -DCMAKE_LIBRARY_PATH=/usr/local/cuda/compat"
fi
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} ${ARM64_DEFS} ${CMAKE_CUDA_DEFS}"
BUILD_DIR="../build/linux/${ARCH}/cuda${CUDA_VARIANT}"