Update Dockerfile for GPU inference

Dockerfile

@@ -1,10 +1,11 @@
-FROM python:3.11-slim
+FROM pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime
 
 WORKDIR /app
 
 # Install system dependencies
 RUN apt-get update && apt-get install -y \
     build-essential \
+    git \
     && rm -rf /var/lib/apt/lists/*
 
 # Copy requirements and install
@@ -14,6 +15,13 @@ RUN pip install --no-cache-dir -r requirements.txt
 # Copy app
 COPY . .
 
+# Create user for HF Space
+RUN useradd -m -u 1000 user
+USER user
+
+# Set HF cache directory
+ENV HF_HOME=/tmp/huggingface
+
 # Expose port
 EXPOSE 7860