Spaces:
Running
Running
Update evaluate_and_submit.py
Browse files- evaluate_and_submit.py +103 -0
evaluate_and_submit.py
CHANGED
|
@@ -0,0 +1,103 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Script to run the agent on Unit 4 questions and optionally submit results to the course scoring API.
|
| 3 |
+
|
| 4 |
+
Usage:
|
| 5 |
+
# Dry run (no submit) on first 5 questions:
|
| 6 |
+
python evaluate_and_submit.py --limit 5
|
| 7 |
+
|
| 8 |
+
# Submit results (requires username and Space URL):
|
| 9 |
+
python evaluate_and_submit.py --submit --username YOUR_HF_USERNAME \\
|
| 10 |
+
--agent-code-url https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE
|
| 11 |
+
|
| 12 |
+
Notes:
|
| 13 |
+
- The scoring API expects exact-match answers, so make sure the agent returns only
|
| 14 |
+
the final answer without extra formatting.
|
| 15 |
+
- Ensure you've run `huggingface-cli login` before submission.
|
| 16 |
+
"""
|
| 17 |
+
import argparse
|
| 18 |
+
import os
|
| 19 |
+
import sys
|
| 20 |
+
import time
|
| 21 |
+
from typing import Any, Dict, List
|
| 22 |
+
|
| 23 |
+
from evaluation_client import ScoringAPIClient
|
| 24 |
+
from code_agent import run_agent
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
def extract_prompt_from_question(q: Dict[str, Any]) -> str:
|
| 28 |
+
"""Extract the actual question/prompt from a question dict."""
|
| 29 |
+
for key in ("question", "prompt", "input", "text", "task"):
|
| 30 |
+
if key in q and isinstance(q[key], str):
|
| 31 |
+
return q[key]
|
| 32 |
+
return str(q)
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
def main(argv: List[str]):
|
| 36 |
+
parser = argparse.ArgumentParser(description="Evaluate agent on course questions and optionally submit.")
|
| 37 |
+
parser.add_argument("--limit", type=int, default=0, help="Max questions to process (0=all)")
|
| 38 |
+
parser.add_argument("--submit", action="store_true", help="Submit answers to scoring API")
|
| 39 |
+
parser.add_argument("--username", type=str, default=os.environ.get("HF_USERNAME"), help="HF username for submission")
|
| 40 |
+
parser.add_argument("--agent-code-url", type=str, default=os.environ.get("AGENT_CODE_URL"), help="Public Space URL for your agent")
|
| 41 |
+
args = parser.parse_args(argv)
|
| 42 |
+
|
| 43 |
+
client = ScoringAPIClient()
|
| 44 |
+
print("Fetching questions from scoring API...")
|
| 45 |
+
questions = client.get_questions()
|
| 46 |
+
if not questions:
|
| 47 |
+
print("ERROR: No questions returned by the API.")
|
| 48 |
+
sys.exit(1)
|
| 49 |
+
|
| 50 |
+
if args.limit > 0:
|
| 51 |
+
questions = questions[:args.limit]
|
| 52 |
+
|
| 53 |
+
print(f"Processing {len(questions)} questions...")
|
| 54 |
+
answers = []
|
| 55 |
+
|
| 56 |
+
for idx, q in enumerate(questions, 1):
|
| 57 |
+
task_id = q.get("task_id") or q.get("id") or q.get("taskId")
|
| 58 |
+
prompt = extract_prompt_from_question(q)
|
| 59 |
+
|
| 60 |
+
print(f"\n[{idx}/{len(questions)}] Task {task_id}")
|
| 61 |
+
print(f" Prompt: {prompt[:100]}...")
|
| 62 |
+
|
| 63 |
+
try:
|
| 64 |
+
ans = run_agent(prompt)
|
| 65 |
+
ans = ans.strip()
|
| 66 |
+
answers.append({"task_id": task_id, "submitted_answer": ans})
|
| 67 |
+
print(f" Answer: {ans[:80]}...")
|
| 68 |
+
except Exception as e:
|
| 69 |
+
print(f" ERROR: {type(e).__name__}: {str(e)[:100]}")
|
| 70 |
+
# Still add an error answer to maintain alignment
|
| 71 |
+
answers.append({"task_id": task_id, "submitted_answer": f"(error) {type(e).__name__}"})
|
| 72 |
+
|
| 73 |
+
# Polite pacing to avoid rate limits
|
| 74 |
+
time.sleep(0.5)
|
| 75 |
+
|
| 76 |
+
print(f"\n✓ Prepared answers for {len(answers)} tasks")
|
| 77 |
+
|
| 78 |
+
if args.submit:
|
| 79 |
+
if not args.username:
|
| 80 |
+
print("ERROR: --submit requires --username (or set HF_USERNAME env var)")
|
| 81 |
+
sys.exit(1)
|
| 82 |
+
if not args.agent_code_url:
|
| 83 |
+
print("ERROR: --submit requires --agent-code-url (or set AGENT_CODE_URL env var)")
|
| 84 |
+
sys.exit(1)
|
| 85 |
+
|
| 86 |
+
print(f"\nSubmitting {len(answers)} answers as user '{args.username}'...")
|
| 87 |
+
print(f"Agent Code URL: {args.agent_code_url}")
|
| 88 |
+
|
| 89 |
+
try:
|
| 90 |
+
resp = client.submit(username=args.username, agent_code=args.agent_code_url, answers=answers)
|
| 91 |
+
print(f"✓ Submission successful!")
|
| 92 |
+
print(f"Response: {resp}")
|
| 93 |
+
except Exception as e:
|
| 94 |
+
print(f"ERROR: Submission failed: {e}")
|
| 95 |
+
sys.exit(1)
|
| 96 |
+
else:
|
| 97 |
+
print("\nDry run complete. To submit, re-run with:")
|
| 98 |
+
print(f" python evaluate_and_submit.py --submit --username YOUR_USERNAME \\")
|
| 99 |
+
print(f" --agent-code-url https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE")
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
if __name__ == "__main__":
|
| 103 |
+
main(sys.argv[1:])
|