File size: 1,761 Bytes
ae0db2a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
"""Answer the example requests with MANTA-4B (one GPU with about 16 GB).

    python example.py                                        # examples/text_request.json and image_request.json
    python example.py --request my_request.json

Prints, per question, the option probabilities (most probable first) and the answer.
"""

from __future__ import annotations

import argparse
import json
import sys
from pathlib import Path

HERE = Path(__file__).resolve().parent
sys.path.insert(0, str(HERE))


def main(argv=None) -> int:
    ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
    ap.add_argument("--request", nargs="*", type=Path,
                    default=[HERE / "examples" / "text_request.json", HERE / "examples" / "image_request.json"])
    ap.add_argument("--release", default=str(HERE), help="the release folder or its Hub repository id")
    ap.add_argument("--device", default="cuda")
    ap.add_argument("--option-order", default="eval", choices=("eval", "given", "sorted"))
    args = ap.parse_args(argv)

    from manta import Manta

    model = Manta.from_pretrained(args.release, device=args.device)
    for path in args.request:
        request = json.loads(path.read_text())
        result = model.predict(request, option_order=args.option_order)
        print(f"\n{path.name} ({result['prompt_tokens']} prompt tokens)")
        for qid, answer in result["questions"].items():
            ranked = sorted(answer["probabilities"].items(), key=lambda kv: -kv[1])
            shown = ", ".join(f"{option} {p:.3f}" for option, p in ranked)
            print(f"  {qid} ({answer['type']}): answer {answer['answer']} | {shown}")
    return 0


if __name__ == "__main__":
    sys.exit(main())