use prebuilt llama.cpp release b10091 (qwen35moe, runtime CPU dispatch) - no compile (no OOM), no -march=native (no SIGILL)

Dockerfile

@@ -1,20 +1,22 @@
 FROM python:3.11-slim
 
+# curl brings libcurl (llama-server links it); libgomp1 for OpenMP.
 RUN apt-get update && apt-get install -y --no-install-recommends \
-    build-essential cmake git curl ca-certificates && rm -rf /var/lib/apt/lists/*
+    curl ca-certificates libgomp1 && rm -rf /var/lib/apt/lists/*
 
-# Build llama.cpp from upstream master — it registers the qwen35moe architecture
-# (hybrid GatedDeltaNet/SSM + MoE + MTP) that POCKET uses. Released llama-cpp-python
-# wheels are too old to load it. Portable CPU build: no -march=native / AVX512, so it
-# never SIGILLs (exit 132) when the runtime CPU differs from the build machine.
-RUN git clone --depth 1 https://github.com/ggml-org/llama.cpp /opt/llama.cpp && \
-    cmake -S /opt/llama.cpp -B /opt/llama.cpp/build \
-        -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF \
-        -DGGML_AVX2=ON -DGGML_FMA=ON -DGGML_F16C=ON -DGGML_AVX512=OFF \
-        -DGGML_CUDA=OFF -DLLAMA_CURL=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_BUILD_TESTS=OFF && \
-    cmake --build /opt/llama.cpp/build --config Release -j 2 --target llama-server && \
-    cp /opt/llama.cpp/build/bin/llama-server /usr/local/bin/llama-server && \
-    rm -rf /opt/llama.cpp
+# Prebuilt llama.cpp (CPU x64). A recent build (b10091) that registers the qwen35moe
+# architecture POCKET uses. It ships runtime CPU-feature dispatch (libggml-cpu-*.so
+# variants), so it's portable — no SIGILL (exit 132) — and needs no source compile,
+# so there is no OOM (exit 137) during the build.
+ARG LLAMA_TAG=b10091
+RUN curl -fsSL -o /tmp/llama.tgz \
+      "https://github.com/ggml-org/llama.cpp/releases/download/${LLAMA_TAG}/llama-${LLAMA_TAG}-bin-ubuntu-x64.tar.gz" && \
+    mkdir -p /opt/llama && tar xzf /tmp/llama.tgz -C /opt/llama && rm /tmp/llama.tgz && \
+    ln -s "/opt/llama/llama-${LLAMA_TAG}" /opt/llama/bin && \
+    /opt/llama/bin/llama-server --version 2>&1 | head -3 || true
+
+ENV LD_LIBRARY_PATH=/opt/llama/bin \
+    PATH="/opt/llama/bin:${PATH}"
 
 WORKDIR /app
 COPY requirements.txt .