Update Dockerfile

Dockerfile

@@ -1,39 +1,45 @@
+# ---- Build stage: 编译 llama.cpp 的 llama-server (CPU only) ----
+FROM python:3.11-slim AS build
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        build-essential cmake git libcurl4-openssl-dev ca-certificates \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN git clone --depth 1 https://github.com/ggml-org/llama.cpp /llama.cpp
+WORKDIR /llama.cpp
+
+# GGML_NATIVE=OFF: HF Space的CPU型号不确定,不能假设AVX512等指令集,
+# 用通用编译避免"illegal instruction"崩溃。LLAMA_CURL=ON 让 -hf 直接从HF拉模型。
+RUN cmake -B build -DGGML_NATIVE=OFF -DLLAMA_CURL=ON -DCMAKE_BUILD_TYPE=Release \
+    && cmake --build build --target llama-server -j "$(nproc)"
+
+# ---- Runtime stage ----
 FROM python:3.11-slim
 
-# 安装系统依赖（含 sox、libsndfile）
-RUN apt-get update && apt-get install -y \
-    sox \
-    libsox-fmt-all \
-    libsndfile1 \
-    ffmpeg \
-    git \
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        libcurl4 ca-certificates \
     && rm -rf /var/lib/apt/lists/*
 
-WORKDIR /app
-
-# 先装 CPU 版 torch（HF 免费 Space 无 GPU）
-RUN pip install --no-cache-dir \
-    torch==2.3.1+cpu \
-    torchaudio==2.3.1+cpu \
-    --index-url https://download.pytorch.org/whl/cpu
-
-# 再装其他依赖
-RUN pip install --no-cache-dir \
-    transformers==4.57.3 \
-    accelerate \
-    einops \
-    gradio \
-    fastapi \
-    uvicorn[standard] \
-    python-multipart \
-    librosa \
-    soundfile \
-    numpy \
-    onnxruntime \
-    qwen-tts
-
-COPY app.py .
+COPY --from=build /llama.cpp/build/bin/llama-server /usr/local/bin/llama-server
+
+# HF Space Docker容器默认以非root用户运行,/tmp通常可写,用它做模型缓存目录
+ENV HF_HOME=/tmp/hf_cache
+RUN mkdir -p /tmp/hf_cache && chmod -R 777 /tmp/hf_cache
+
+# ---- 可在Space的 Settings > Variables 里覆盖,无需重新构建镜像 ----
+ENV MODEL_REPO="Qwen/Qwen3-4B-Instruct-2507-GGUF"
+ENV MODEL_QUANT="Q4_K_M"
+ENV CTX_SIZE="4096"
+ENV N_THREADS="2"
 
 EXPOSE 7860
 
-CMD ["python", "app.py"]
+# --jinja: 启用chat template里的tool-calling语法解析,让 /v1/chat/completions
+#          支持传 tools=[...] 并正确输出 tool_calls
+CMD ["sh", "-c", "llama-server \
+    -hf ${MODEL_REPO}:${MODEL_QUANT} \
+    --host 0.0.0.0 \
+    --port 7860 \
+    -c ${CTX_SIZE} \
+    -t ${N_THREADS} \
+    --jinja"]