scripts/smoke_test.py

"""Smoke-test the oracle against whatever provider is configured (LM Studio).

Generates a couple of moody synthetic images + a fragment, runs the full
interpret() flow, and pretty-prints the result. Lets us judge prompt/model
quality before any UI exists.

Usage:
    .venv/bin/python -m scripts.smoke_test [model]
e.g.
    .venv/bin/python -m scripts.smoke_test google/gemma-4-12b
"""

import base64
import io
import json
import os
import sys
import time

from PIL import Image, ImageDraw
from dotenv import load_dotenv

load_dotenv()

# Allow overriding the model from the CLI for quick A/B (e4b vs 12b).
if len(sys.argv) > 1:
    os.environ["CHALCHITRA_MODEL"] = sys.argv[1]

from backend import interpret  # noqa: E402  (after env setup)
from backend.providers import get_provider  # noqa: E402


def moody_jpeg(top, bottom, blob=None) -> str:
    """A vertical gradient with an optional soft blob — enough texture for a read."""
    w, h = 640, 480
    img = Image.new("RGB", (w, h))
    px = img.load()
    for y in range(h):
        t = y / h
        px_row = tuple(int(top[i] * (1 - t) + bottom[i] * t) for i in range(3))
        for x in range(w):
            px[x, y] = px_row
    if blob:
        d = ImageDraw.Draw(img, "RGBA")
        d.ellipse([w * 0.5, h * 0.25, w * 0.85, h * 0.6], fill=blob)
    buf = io.BytesIO()
    img.save(buf, format="JPEG", quality=72)
    b64 = base64.b64encode(buf.getvalue()).decode()
    return f"data:image/jpeg;base64,{b64}"


def main() -> None:
    p = get_provider()
    print(f"provider={p.name}  model={getattr(p, 'model', '?')}  "
          f"base_url={getattr(p, 'base_url', 'n/a')}")

    images = [
        moody_jpeg((20, 24, 40), (90, 70, 60), blob=(255, 200, 120, 60)),   # dusk / amber window
        moody_jpeg((12, 14, 18), (40, 50, 70)),                              # blue night
    ]
    fragment = "rain on the window, the city out of focus"

    print(f"\nfragment: {fragment!r}\nimages: {len(images)} synthetic mood frames\n")
    t0 = time.time()
    result = interpret(images, fragment)
    dt = time.time() - t0

    print(json.dumps(result, indent=2, ensure_ascii=False))
    print(f"\n--- {dt:.1f}s ---")
    print(f"interpretation: {len(result['interpretation'])} chars, "
          f"{len(result['films'])} films")


if __name__ == "__main__":
    main()