Sync with upstream ggml-org/llama.cpp tag b7751

Dockerfile

@@ -0,0 +1,19 @@
+FROM image.sourcefind.cn:5000/dcu/admin/base/dtk:25.04.2-ubuntu22.04-py3.10 as builder
+ADD . /src
+WORKDIR /src
+RUN mkdir -p /opt/dtk-25.04.2/llvm/lib/clang/lib/linux && cp /opt/dtk-25.04.2/llvm/lib/clang/17.0.0/lib/linux/*  /opt/dtk-25.04.2/llvm/lib/clang/lib/linux/ -r 
+RUN --mount=type=bind,from=hyhal,source=/,target=/opt/hyhal,readonly \
+ HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
+ cmake -S . -B build \
+  -DGGML_HIP=ON \
+  -DAMDGPU_TARGETS=gfx928 \
+  -DGGML_CUDA_FORCE_MMQ=ON \
+  -DGGML_CUDA_FA=OFF \
+  -DGGML_HIP_MMQ_MFMA=OFF \
+  -DCMAKE_BUILD_TYPE=Release &&\
+   cmake --build build -j
+
+FROM image.sourcefind.cn:5000/dcu/admin/base/dtk:25.04.2-ubuntu22.04-py3.10
+COPY --from=builder /src/build/bin /app
+WORKDIR /app
+ENTRYPOINT [ "/app/llama-cli" ]