| | import requests |
| | import pandas as pd |
| | |
| | results = [] |
| | out_path = "/home/data/result/test.parquet" |
| | df = pd.read_parquet("/home/data/raw/test/4201_2355_full_label_1000-8192.parquet") |
| | |
| | url = "http://localhost:5000/get_reward" |
| | total=0 |
| | correct = 0 |
| |
|
| | for idx, row in df.iterrows(): |
| | |
| | q1 = str(row["chosen_prompt"]) + str(row["chosen"]) |
| | q2 = str(row["chosen_prompt"]) + str(row["reject"]) |
| | print(q1) |
| | print("\n") |
| | print(q2) |
| | payload = {"query": [q1, q2]} |
| |
|
| | try: |
| | |
| | response = requests.post(url, json=payload) |
| | scores= response.json().get("rewards", []) |
| | score1, score2 = scores[0], scores[1] |
| | right = 1 if score1 > score2 else 0 |
| | total += 1 |
| | if score1 > score2: |
| | correct += 1 |
| | scores= response.json().get("rewards", []) |
| | acc = correct / total * 100 |
| | print(f"Row {idx}: score1={score1:.4f}, score2={score2:.4f}, " |
| | f"Correct={score1 > score2}, RunningAcc={acc:.2f}%") |
| | results.append({ |
| | "q1": q1, |
| | "q2": q2, |
| | "chosen_score": score1, |
| | "reject_score": score2, |
| | "right": right, |
| | "chosen_label": row["chosen_label"], |
| | "chosen_violations": row["chosen_violations"], |
| | "reject_label": row["reject_label"], |
| | "reject_violations": row["reject_violations"] |
| | }) |
| |
|
| | except Exception as e: |
| | print(f"Row {idx} 出错:", e) |
| | if total >= 2: |
| | break |
| | results_df = pd.DataFrame(results) |
| | results_df.to_parquet(out_path) |