File size: 1,937 Bytes
e5f4c09 0cc1b2b e5f4c09 | 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 46 47 48 49 50 51 52 53 54 55 56 57 58 | #!/usr/bin/env python3
"""Run batched AspectBench predictions from JSON or JSONL."""
from __future__ import annotations
import argparse
from pathlib import Path
from inference import InferenceEngine, load_records, write_json
from model_registry import LANGUAGES, MODES, MODEL_SPECS
SCRIPT_DIR = Path(__file__).resolve().parent
DEFAULT_MODEL_ROOT = SCRIPT_DIR.parent / "models"
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(
description="Predict a JSON/JSONL list whose articles contain <aspect>...</aspect>."
)
parser.add_argument("--model-name", required=True, choices=sorted(MODEL_SPECS))
parser.add_argument("--language", required=True, choices=LANGUAGES)
parser.add_argument("--mode", required=True, choices=MODES)
parser.add_argument("--input", required=True, type=Path)
parser.add_argument("--output", type=Path)
parser.add_argument("--batch-size", type=int, default=8)
parser.add_argument("--model-root", type=Path, default=DEFAULT_MODEL_ROOT)
parser.add_argument("--base-model-root", type=Path)
parser.add_argument("--device", default="auto")
parser.add_argument("--mc-passes", type=int, default=0)
parser.add_argument("--seed", type=int, default=42)
return parser.parse_args()
def main() -> None:
args = parse_args()
engine = InferenceEngine(
model_name=args.model_name,
language=args.language,
mode=args.mode,
model_root=args.model_root,
base_model_root=args.base_model_root,
device=args.device,
)
predictions = engine.predict_batch(
load_records(args.input),
batch_size=args.batch_size,
mc_passes=args.mc_passes,
seed=args.seed,
)
write_json(predictions, args.output)
if args.output is not None:
print(f"Wrote {len(predictions)} predictions to {args.output.resolve()}")
if __name__ == "__main__":
main()
|