"""Evaluate labeled local audio without a server. Built with Qwen; see LICENSE."""
import argparse
import json
from pathlib import Path
from toddler_jev import ToddlerJev
from train import read_data

def main():
    p=argparse.ArgumentParser()
    p.add_argument('--data',required=True)
    p.add_argument('--output',required=True)
    p.add_argument('--batch-size',type=int,default=1)
    p.add_argument('--model',default=str(Path(__file__).parent))
    p.add_argument('--revision',default='v1.0.0')
    p.add_argument('--device',default=None)
    args=p.parse_args()
    if args.batch_size<1:p.error('Batch size must be positive')
    target=Path(args.output)
    if target.exists():p.error('Output must be a new file')
    rows=read_data(args.data)
    model=ToddlerJev.from_pretrained(args.model,revision=args.revision,device=args.device)
    correct=0
    with target.open('x',encoding='utf-8') as f:
        for i in range(0,len(rows),args.batch_size):
            batch=rows[i:i+args.batch_size]
            for row,result in zip(batch,model.predict_batch(batch)):
                result['id']=row.get('id',i)
                result['target_index']=row['target']
                result['correct']=result['decision_index']==row['target']
                correct+=result['correct']
                f.write(json.dumps(result)+'\n')
            f.flush()
    print(json.dumps({'examples':len(rows),'correct':correct,'accuracy':correct/len(rows)}))

if __name__=='__main__':main()