docker/docker-entrypoint.sh

#!/usr/bin/env bash
# Container entrypoint: ensures the model is present, then launches the Gradio app.
#
# Behavior:
#   - If MODEL_SKIP_DOWNLOAD=1, skip the download step (assume the model is
#     already mounted via a volume or baked into the image).
#   - Otherwise run download_model.py to fetch weights + config into MODEL_DIR.
#   - Then exec the Gradio app (default) or whatever command was passed.
set -euo pipefail

MODEL_DIR="${MODEL_DIR:-/data/model}"
export MODEL_DIR

echo "[entrypoint] MODEL_DIR=${MODEL_DIR}"
echo "[entrypoint] DEVICE=$([[ -e /dev/nvidia0 ]] && echo cuda || echo cpu)"

if [[ "${MODEL_SKIP_DOWNLOAD:-0}" != "1" ]]; then
    if [[ -f "${MODEL_DIR}/model.safetensors" && -f "${MODEL_DIR}/config.json" ]]; then
        echo "[entrypoint] model already present, skipping download"
    else
        echo "[entrypoint] downloading model artifacts..."
        python /app/download_model.py --model-dir "${MODEL_DIR}"
    fi
else
    echo "[entrypoint] MODEL_SKIP_DOWNLOAD=1, skipping download"
    if [[ ! -f "${MODEL_DIR}/model.safetensors" ]]; then
        echo "[entrypoint] WARNING: ${MODEL_DIR}/model.safetensors missing — app will fail to load" >&2
    fi
fi

if [[ $# -eq 0 ]]; then
    echo "[entrypoint] launching Gradio app on ${GRADIO_SERVER_NAME:-0.0.0.0}:${GRADIO_SERVER_PORT:-7860}"
    exec python /app/app.py
fi

exec "$@"