"""JSON-lines inference with local pinned E5. Does not execute suggested tools.""" import argparse import json from release_runtime import load_predictor parser = argparse.ArgumentParser() parser.add_argument("--encoder", required=True, help="Local pinned multilingual-e5-small directory") parser.add_argument("--input", required=True, help="JSONL native requests or published dataset rows") args = parser.parse_args() predictor = load_predictor(args.encoder) with open(args.input, encoding="utf-8") as handle: for line in handle: if line.strip(): row = json.loads(line) request = json.loads(row["request_json"]) if "request_json" in row else row result = predictor.predict(request) result.pop("model", None) print(json.dumps(result, ensure_ascii=False, allow_nan=False), flush=True)