using ollama

Dockerfile

@@ -1,34 +1,26 @@
-# Use a standard Python 3.12 image
-FROM python:3.12-slim
+# --- 1. Use the official Ollama pre-built image ---
+FROM ollama/ollama
 
-WORKDIR /app
-
-# --- 1. Install build-essential and cmake ---
-# This is necessary for compiling the C++ code
-RUN apt-get update && apt-get install -y build-essential cmake
-
-# --- 2. Install Python Dependencies (with CPU-only build) ---
-# We set CMAKE_ARGS to disable CUDA, which makes the
-# build *much* faster and avoids the job timeout.
-ENV CMAKE_ARGS="-DLLAMA_CUDA=OFF"
-RUN pip install "llama-cpp-python[server]" huggingface_hub
+# --- 2. Install Python & dependencies for our download script ---
+# The base image is Debian, so we can use apt-get
+RUN apt-get update && apt-get install -y python3 python3-pip
+RUN pip install huggingface_hub
 
-# --- 3. Model Download ---
-# This part is correct and remains the same.
+# --- 3. Download the GGUF model ---
+WORKDIR /app
 COPY download_model.py .
 ARG HF_TOKEN
 RUN --mount=type=secret,id=HF_TOKEN \
-    python download_model.py
+    python3 download_model.py
+
+# --- 4. Create the Ollama "Modelfile" ---
+# This file tells Ollama to use our downloaded GGUF
+RUN echo "FROM /app/prem-1B-SQL.Q8_0.gguf" > /app/Modelfile
+
+# --- 5. Import the model into Ollama's registry ---
+# This makes the model available to serve
+RUN ollama create prem-sql-api -f /app/Modelfile
 
-# --- 4. Server Runtime ---
-# This part is also correct and remains the same.
-EXPOSE 8000
-CMD [ \
-    "python", \
-    "-m", "llama_cpp.server", \
-    "--model", "prem-1B-SQL.Q8_0.gguf", \
-    "--n_gpu_layers", "0", \
-    "--port", "8000", \
-    "--host", "0.0.0.0", \
-    "--api_key_env_var", "API_KEY" \
-]
+# The base image's default command is "ollama serve",
+# which will automatically start the API server on port 11434.
+# It will also serve our newly created "prem-sql-api" model.