Updating toolboxes with a two-stage build process to reduce size

toolboxes/Dockerfile.rocm-6.4.2-rocwmma

@@ -1,45 +1,65 @@
-FROM fedora:rawhide
+# build
+FROM registry.fedoraproject.org/fedora:rawhide AS builder
 
-# Install build dependencies and tools
-RUN dnf install -y \
+# deps + rocm toolchain
+RUN dnf -y --nodocs --setopt=install_weak_deps=False install \
        make gcc cmake lld clang clang-devel compiler-rt libcurl-devel \
-       rocminfo radeontop 'rocm-*' 'rocblas-*' 'hipblas' 'hipblas-*' \
-       git vim rsync \
-    && dnf clean all
+       rocminfo radeontop 'rocm-*' 'rocblas-*' hipblas 'hipblas-*' \
+       git vim rsync sudo tar xz \
+    && dnf clean all && rm -rf /var/cache/dnf/*
 
-
-WORKDIR /opt/
+# rocWMMA headers
+WORKDIR /opt
 RUN git clone -b release/rocm-rel-7.0 https://github.com/ROCm/rocWMMA.git
 RUN sudo mkdir -p /usr/include/rocwmma
 RUN sudo rsync -a rocWMMA/library/include/rocwmma/ /usr/include/rocwmma/
 
-# Set up working directory
+# llama.cpp
 WORKDIR /opt/llama.cpp
-
-# Clone llama.cpp repository (with submodules)
 RUN git clone --recursive https://github.com/ggerganov/llama.cpp.git .
 
-# Build llama.cpp with HIP support
+# build + install
 RUN git clean -xdf \
-    && git pull \
-    && git submodule update --recursive \
-    && \
-    # Configure and compile with HIP toolchain
-    HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
-      cmake -S . -B build \
-            -DGGML_HIP=ON \
-            -DAMDGPU_TARGETS=gfx1151 \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DLLAMA_HIP_UMA=ON \
-            -DGGML_HIP_ROCWMMA_FATTN=ON \
-    && cmake --build build --config Release -- -j$(nproc) \
-    && cmake --install build --config Release
+ && git pull \
+ && git submodule update --recursive \
+ && HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
+    cmake -S . -B build \
+      -DGGML_HIP=ON \
+      -DAMDGPU_TARGETS=gfx1151 \
+      -DCMAKE_BUILD_TYPE=Release \
+      -DLLAMA_HIP_UMA=ON \
+      -DGGML_HIP_ROCWMMA_FATTN=ON \
+ && cmake --build build --config Release -- -j$(nproc) \
+ && cmake --install build --config Release
 
+# make ld see libllama in builder too (kept same step)
 RUN find /opt/llama.cpp/build -type f -name 'lib*.so*' -exec cp {} /usr/lib64/ \; \
  && ldconfig
 
+
+# runtime
+FROM registry.fedoraproject.org/fedora-minimal:rawhide
+
+# runtime deps (same rocm packages; no build toolchain)
+RUN microdnf -y --nodocs --setopt=install_weak_deps=0 install \
+      bash ca-certificates libatomic libstdc++ libgcc \
+      rocminfo radeontop 'rocm-*' 'rocblas-*' hipblas 'hipblas-*' \
+  && microdnf clean all && rm -rf /var/cache/dnf/*
+
+# bits from builder
+COPY --from=builder /usr/local/ /usr/local/
+COPY --from=builder /usr/include/rocwmma /usr/include/rocwmma
+
+# ensure libllama is on the linker path
+RUN echo "/usr/local/lib"  > /etc/ld.so.conf.d/local.conf \
+ && echo "/usr/local/lib64" >> /etc/ld.so.conf.d/local.conf \
+ && ldconfig \
+ && cp -n /usr/local/lib/libllama*.so* /usr/lib64/ 2>/dev/null || true \
+ && ldconfig
+
+# helper
 COPY gguf-vram-estimator.py /usr/local/bin/gguf-vram-estimator.py
 RUN chmod +x /usr/local/bin/gguf-vram-estimator.py
 
-# Default to interactive shell
+# shell
 CMD ["/bin/bash"]