Configure for CPU-only zero-gpu deployment

Dockerfile

@@ -2,7 +2,7 @@ FROM python:3.12-slim
 
 WORKDIR /app
 
-# System dependencies for audio and Python packages
+# System dependencies (CPU-only, no CUDA)
 RUN apt-get update && apt-get install -y --no-install-recommends \
     build-essential \
     cmake \
@@ -15,10 +15,11 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 # Copy project files
 COPY . /app/
 
-# Install venv and dependencies
-RUN python3 -m venv /opt/venv
-ENV PATH="/opt/venv/bin:$PATH"
+# Install dependencies
 RUN pip install --upgrade pip setuptools wheel
+# Install llama-cpp-python without GPU support (CPU-only)
+RUN pip install --no-cache-dir llama-cpp-python --no-build-isolation
+# Install remaining requirements
 RUN pip install --no-cache-dir -r requirements.txt
 
 EXPOSE 7860

config.py

@@ -21,13 +21,13 @@ HF_TOKEN = os.environ.get("HF_TOKEN", "")
 # ASR: Hugging Face Moonshine
 TRANSCRIBE_MODEL_ID = "UsefulSensors/moonshine-tiny"
 TRANSCRIBE_LANGUAGE = "en"          # Moonshine Tiny is English ASR
-TRANSCRIBE_DEVICE = "cuda:0"        # falls back to "cpu" automatically in code
+TRANSCRIBE_DEVICE = "cpu"           # CPU-only for zero-gpu deployment
 
 # Intent Parser: Qwen2.5-7B-Instruct (GGUF via llama-cpp-python)
-# Using q3_k_m quantization (3.55 GB, good quality/size tradeoff).
-# Already downloaded and available in ./models/
-QWEN_GGUF_PATH = MODELS_DIR / "qwen2.5-7b-instruct-q3_k_m.gguf"
-QWEN_N_GPU_LAYERS = 20     # offload 20 transformer layers to GPU (~0.8 GB VRAM)
+# Using q2_k quantization for smaller size on CPU (~2.3 GB).
+# Download from HF Hub on first run if not cached.
+QWEN_GGUF_PATH = MODELS_DIR / "qwen2.5-7b-instruct-q2_k.gguf"
+QWEN_N_GPU_LAYERS = 0      # CPU-only: no GPU layer offloading
 QWEN_N_CTX = 4096          # context window sufficient for a call transcript
 QWEN_MAX_TOKENS = 512      # max tokens for the structured JSON response
 QWEN_TEMPERATURE = 0.1     # near-deterministic for structured output