| """Run the assignment agent locally without submitting results.""" |
|
|
| import argparse |
| import textwrap |
|
|
| import requests |
|
|
| from agent import BasicAgent |
|
|
|
|
| DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space" |
|
|
|
|
| def main() -> None: |
| parser = argparse.ArgumentParser(description=__doc__) |
| parser.add_argument( |
| "--limit", |
| type=int, |
| default=None, |
| help="Evaluate only the first N questions.", |
| ) |
| parser.add_argument( |
| "--start", |
| type=int, |
| default=0, |
| help="Skip the first N questions.", |
| ) |
| parser.add_argument( |
| "--task-id", |
| action="append", |
| dest="task_ids", |
| help="Evaluate only these task ids. Repeatable.", |
| ) |
| parser.add_argument( |
| "--model", |
| default=None, |
| help="Model id to use, e.g. anthropic/claude-opus-4-1. Defaults to the configured model.", |
| ) |
| args = parser.parse_args() |
|
|
| response = requests.get(f"{DEFAULT_API_URL}/questions", timeout=15) |
| response.raise_for_status() |
| questions = response.json() |
| if args.task_ids: |
| questions = [q for q in questions if q.get("task_id") in set(args.task_ids)] |
| else: |
| questions = questions[args.start:] |
| if args.limit is not None: |
| questions = questions[:args.limit] |
|
|
| print("\n" + "=" * 72) |
| print(f"LOCAL EVALUATION | {len(questions)} question(s)") |
| print("Nothing will be submitted.") |
| print("=" * 72) |
| agent = BasicAgent(verbose=True, model_id=args.model) |
|
|
| for index, item in enumerate(questions, start=1): |
| task_id = item.get("task_id") |
| question = item.get("question") |
| if not task_id or question is None: |
| print(f"\n--- Question {index} | SKIPPED ---") |
| print(f"Malformed question: {item}") |
| continue |
|
|
| print(f"\n--- Question {index} | {task_id} | TRACE ---") |
| print("Question:") |
| print(textwrap.indent(question, " ")) |
| print("Loop:") |
| try: |
| answer = agent(question, task_id=task_id, file_name=item.get("file_name")) |
| except Exception as error: |
| print(f"\n--- Question {index} | {task_id} | ERROR ---") |
| print(f"{error}") |
| continue |
|
|
| print(f"\n--- Question {index} | {task_id} ---") |
| print("Answer:") |
| print(textwrap.indent(answer, " ")) |
|
|
| print("\n" + "=" * 72) |
| print("LOCAL EVALUATION COMPLETE") |
| print("=" * 72) |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
