Update Dockerfile

Dockerfile

@@ -1,54 +1,50 @@
-# Use the official Ollama base image, which already includes Ollama
-FROM ollama/ollama:latest
+# Start with a Python base image, which is convenient for running the Gradio app.
+FROM python:3.9
 
 # Set environment variables to prevent interactive prompts during apt operations
 ENV DEBIAN_FRONTEND=noninteractive
 
-# Install system dependencies: git (optional), python3 and pip for the application
-RUN apt update && apt install -y git python3 python3-pip
+# Install system dependencies: curl for Ollama and wget for model download.
+RUN apt-get update && apt-get install -y curl wget
 
-# Set up a non-root user as recommended for Hugging Face Spaces
+# Install Ollama using its official installation script.
+RUN curl -fsSL https://ollama.com/install.sh | sh
+
+# Set environment variables for the model for easy modification.
+ENV MODEL_REPO="unsloth/gemma-3-4b-it-qat-GGUF"
+ENV MODEL_FILENAME="gemma-3-4b-it-qat.Q4_K_M.gguf"
+
+# Create a non-root user and switch to its home directory. This is a best practice for security and avoids permission errors with model caching.
+# As seen in the Hugging Face Docker guide. [1]
 RUN useradd -m -u 1000 user
-USER user
-ENV HOME=/home/user \
-	PATH=/home/user/.local/bin:$PATH \
-    OLLAMA_HOST=0.0.0.0 # Allow Ollama to be accessed from outside localhost
-
-# Set the working directory inside the container
-WORKDIR $HOME/app
-
-# Define the model to be pulled
-ENV OLLAMA_HF_MODEL="hf.co/unsloth/gemma-3-4b-it-qat-GGUF:Q4_K_M"
-
-# Pull the model during the build process
-# Start ollama serve in background, wait for it, pull the model, then kill ollama.
-# This ensures the model is downloaded and cached within the image.
-RUN ollama serve & \
-    echo "Waiting for Ollama to start for model pull..." && \
-    sleep 5 && \
-    while ! curl -s http://localhost:11434 > /dev/null; do \
-        sleep 1; \
-    done && \
-    echo "Ollama started. Pulling model: ${OLLAMA_HF_MODEL}" && \
-    ollama pull ${OLLAMA_HF_MODEL} && \
-    echo "Model pull complete. Stopping Ollama for build process." && \
-    pkill ollama || true # '|| true' to prevent build failure if pkill returns non-zero when ollama already stopped
-
-# Copy the application files
-COPY app.py .
-COPY requirements.txt .
-COPY run.sh .
-
-# Install Python dependencies required by your Gradio application
+WORKDIR /home/user/app
+
+# Download the GGUF model file directly from Hugging Face Hub.
+RUN wget -O ${MODEL_FILENAME} https://huggingface.co/${MODEL_REPO}/resolve/main/${MODEL_FILENAME}
+
+# Copy the application files and set the correct ownership to the new user. [1]
+COPY --chown=user:user Modelfile .
+COPY --chown=user:user app.py .
+COPY --chown=user:user requirements.txt .
+COPY --chown=user:user run.sh .
+
+# Install Python dependencies.
 RUN pip install --no-cache-dir -r requirements.txt
 
-# Make the startup script executable
+# Make the startup script executable.
 RUN chmod +x run.sh
 
-# Expose the port that your Gradio application will listen on.
-# Hugging Face Spaces typically use port 7860 for Gradio apps.
+# Switch to the non-root user. [1]
+USER user
+
+# Set environment variables for the user and for Ollama.
+# This ensures cached models are stored in a writable directory and that Ollama is accessible within the container network. [1]
+ENV HOME=/home/user
+ENV PATH=/home/user/.local/bin:$PATH
+ENV OLLAMA_HOST=0.0.0.0
+
+# Expose the port for the Gradio application.
 EXPOSE 7860
 
-# Set the entrypoint for the container to execute our startup script.
-# This script will start Ollama and then your application.
+# Set the command to run our startup script.
 CMD ["./run.sh"]