#!/usr/bin/env bash
# onw - Ore no NPU ga konna ni ugoku wake nai: one-command start (Linux).
#   ./start.sh                                   Gemma 4 E4B (default: text + image, fits 16 GB machines)
#   ./start.sh ryugyosoft/Qwen3.6-35B-A3B-onw    any onw model: a Hugging Face repo id or a local folder
#   ./start.sh MODEL --port 8080 --host 0.0.0.0  further options go to "onw serve"
# First run: creates .venv (~1 GB), downloads the model, compiles it for the NPU. Later runs start in ~20 s.
# Afterwards: source .venv/bin/activate, then onw serve / onw chat / onw convert.
# Without the Intel NPU driver everything runs on the CPU (slowly).
set -e
cd "$(dirname "$0")"
export PYTHONUTF8=1
MODEL="ryugyosoft/gemma-4-E4B-it-onw"
if [ $# -gt 0 ] && [ "${1#-}" = "$1" ]; then MODEL="$1"; shift; fi

if ! python3 -c "import sys; sys.exit(0 if sys.version_info >= (3, 10) else 1)" 2>/dev/null; then
    echo "Python 3.10 or newer is needed (Ubuntu 22.04 and later ship it):  sudo apt install python3"
    exit 1
fi

if [ ! -f .venv/installed.txt ]; then
    echo "[setup] creating Python environment in .venv ..."
    if ! python3 -m venv .venv; then
        echo "python3-venv is missing:  sudo apt install python3-venv"
        exit 1
    fi
    .venv/bin/python -m pip install "pip>=26.2.2"
    .venv/bin/python -m pip install -e . \
        --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/pre-release
    echo ok > .venv/installed.txt
fi

echo "[start] $MODEL (first time: download + NPU compile, please wait) ..."
exec .venv/bin/onw serve "$MODEL" --open "$@"