Update Dockerfile

Dockerfile

@@ -1,10 +1,57 @@
-FROM ghcr.io/ggml-org/llama.cpp:server
+FROM debian:bookworm-slim AS builder
 
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    cmake \
+    git \
+    curl \
+    libcurl4-openssl-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /build
+
+# Pin a known-good release; bump as needed
+RUN git clone --depth 1 --branch b3600 https://github.com/ggerganov/llama.cpp.git
+
+WORKDIR /build/llama.cpp
+
+# Build with CPU-only, no CUDA. Enable native arch optimizations (AVX2 etc if host supports it)
+RUN cmake -B build \
+    -DGGML_NATIVE=ON \
+    -DGGML_CUDA=OFF \
+    -DLLAMA_CURL=ON \
+    -DCMAKE_BUILD_TYPE=Release \
+    && cmake --build build --config Release -j2 --target llama-server
+
+# ---- Runtime stage ----
+FROM debian:bookworm-slim
+
+RUN apt-get update && apt-get install -y \
+    libcurl4 \
+    ca-certificates \
+    curl \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+
+COPY --from=builder /build/llama.cpp/build/bin/llama-server /app/llama-server
+
+# Pull the GGUF model at build time (Q4_K_M quant, ~1GB for 1.5B model)
+# Swap this URL for whichever quant repo you land on (e.g. bartowski/Qwen2.5-1.5B-Instruct-GGUF)
+ARG MODEL_URL=https://huggingface.co/bartowski/Qwen2.5-1.5B-Instruct-GGUF/resolve/main/Qwen2.5-1.5B-Instruct-Q4_K_M.gguf
+RUN curl -L -o /app/model.gguf "${MODEL_URL}"
+
+# HF Spaces expects the app to listen on port 7860
 EXPOSE 7860
 
-CMD ["--hf-repo", "Qwen/Qwen2.5-3B-Instruct-GGUF", \
-     "--hf-file", "qwen2.5-3b-instruct-q4_k_m.gguf", \
-     "--port", "7860", \
+# --threads 2 to match vCPU count, avoid oversubscription
+# --ctx-size kept small since bot commands are short
+# -ngl 0 forces CPU-only (no GPU layers offloaded)
+CMD ["/app/llama-server", \
+     "-m", "/app/model.gguf", \
      "--host", "0.0.0.0", \
-     "--chat-template", "chatml", \
-     "-c", "8192"]
+     "--port", "7860", \
+     "--threads", "2", \
+     "--ctx-size", "2048", \
+     "-ngl", "0", \
+     "--batch-size", "512"]