scripts/ab_live.py

"""A/B the baseline vs refined prompt against the LIVE Space (Qwen2.5-VL).

Encodes local image files and POSTs each set to /api/interpret twice — once with
the baseline prompt, once refined — using the per-request `refined` override, so
both variants see the exact same image on the same model. Prints them to compare.

Usage:
    .venv/bin/python -m scripts.ab_live IMG [IMG ...] [--fragment "a line"]

Up to 4 images make one "moment".
"""

import argparse
import base64
import mimetypes
import sys
import time

import httpx

SPACE = "https://build-small-hackathon-chalchitra.hf.space/api/interpret"


def to_data_url(path: str) -> str:
    mime = mimetypes.guess_type(path)[0] or "image/jpeg"
    with open(path, "rb") as f:
        return f"data:{mime};base64," + base64.b64encode(f.read()).decode()


def run(images, fragment, refined):
    body = {"images": images, "fragment": fragment, "refined": refined}
    t0 = time.time()
    r = httpx.post(SPACE, json=body, timeout=600)
    dt = time.time() - t0
    if r.status_code != 200:
        return None, dt, f"HTTP {r.status_code}: {r.text[:200]}"
    return r.json(), dt, None


def show(label, data, dt, err):
    print(f"\n{'='*74}\n{label}   [{dt:.1f}s]\n{'-'*74}")
    if err:
        print(f"  ERROR: {err}")
        return
    print(data["interpretation"])
    print("\nfilms: " + " · ".join(f"{f['title']} ({f['year']})" for f in data["films"]))


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("images", nargs="+")
    ap.add_argument("--fragment", default="")
    args = ap.parse_args()

    images = [to_data_url(p) for p in args.images[:4]]
    print(f"{len(images)} image(s); fragment={args.fragment!r}\nlive: {SPACE}")
    for refined in (False, True):
        data, dt, err = run(images, args.fragment, refined)
        show("REFINED" if refined else "BASELINE", data, dt, err)


if __name__ == "__main__":
    main()