Optimize Docker build for faster HF Space startup

Dockerfile

@@ -1,21 +1,25 @@
-FROM python:3.11-slim
+FROM python:3.10-slim
 
 ENV PYTHONDONTWRITEBYTECODE=1 \
     PYTHONUNBUFFERED=1 \
     PIP_NO_CACHE_DIR=1 \
-    CMAKE_ARGS="-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS"
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_DEFAULT_TIMEOUT=120 \
+    HF_HOME=/tmp/hf-cache \
+    HF_HUB_ENABLE_HF_TRANSFER=1
 
 RUN apt-get update && apt-get install -y --no-install-recommends \
-    build-essential \
-    cmake \
-    pkg-config \
-    libopenblas-dev \
+    ca-certificates \
+    curl \
+    git \
+    libgomp1 \
   && rm -rf /var/lib/apt/lists/*
 
 WORKDIR /app
 
 COPY requirements.txt /app/requirements.txt
-RUN pip install --upgrade pip && pip install -r /app/requirements.txt
+RUN python -m pip install --upgrade pip setuptools wheel && \
+    pip install --prefer-binary -r /app/requirements.txt
 
 COPY start_server.py /app/start_server.py
 
@@ -23,8 +27,8 @@ ENV PORT=7860 \
     MODEL_REPO=unsloth/Qwen3.5-0.8B-GGUF \
     MODEL_FILE=Qwen3.5-0.8B-Q4_K_M.gguf \
     MODEL_DIR=/tmp/models \
-    N_CTX=4096 \
-    N_THREADS=4 \
+    N_CTX=2048 \
+    N_THREADS=2 \
     CHAT_FORMAT=chatml
 
 EXPOSE 7860

README.md

@@ -5,6 +5,7 @@ colorFrom: yellow
 colorTo: gray
 sdk: docker
 app_port: 7860
+startup_duration_timeout: 2h
 pinned: false
 ---
 
@@ -39,8 +40,8 @@ Recommended defaults:
 
 - `MODEL_REPO=unsloth/Qwen3.5-0.8B-GGUF`
 - `MODEL_FILE=Qwen3.5-0.8B-Q4_K_M.gguf`
-- `N_CTX=4096`
-- `N_THREADS=4`
+- `N_CTX=2048`
+- `N_THREADS=2`
 - `CHAT_FORMAT=chatml`
 
 Optional:

requirements.txt

@@ -1,2 +1,3 @@
 llama-cpp-python[server]>=0.2.90
 huggingface_hub>=0.25.0
+hf_transfer>=0.1.8