Spaces:
Running
Running
| import json | |
| import logging | |
| import os | |
| import subprocess | |
| import time | |
| import pandas as pd | |
| from huggingface_hub import snapshot_download | |
| from src.envs import EVAL_RESULTS_PATH | |
| # Configure logging | |
| logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s") | |
| def time_diff_wrapper(func): | |
| def wrapper(*args, **kwargs): | |
| start_time = time.time() | |
| result = func(*args, **kwargs) | |
| end_time = time.time() | |
| diff = end_time - start_time | |
| logging.info("Time taken for %s: %s seconds", func.__name__, diff) | |
| return result | |
| return wrapper | |
| def download_dataset(repo_id, local_dir, repo_type="dataset", max_attempts=3, backoff_factor=1.5): | |
| """Download dataset with exponential backoff retries.""" | |
| attempt = 0 | |
| while attempt < max_attempts: | |
| try: | |
| logging.info("Downloading %s to %s", repo_id, local_dir) | |
| snapshot_download( | |
| repo_id=repo_id, | |
| local_dir=local_dir, | |
| repo_type=repo_type, | |
| tqdm_class=None, | |
| token=os.environ.get("HF_TOKEN_PRIVATE"), | |
| etag_timeout=30, | |
| max_workers=8, | |
| ) | |
| logging.info("Download successful") | |
| return | |
| except Exception as e: | |
| wait_time = backoff_factor**attempt | |
| logging.error("Error downloading %s: %s, retrying in %ss", repo_id, e, wait_time) | |
| time.sleep(wait_time) | |
| attempt += 1 | |
| logging.error("Failed to download %s after %s attempts", repo_id, max_attempts) | |
| def build_leadearboard_df(): | |
| """Initializes the application space, loading only necessary data.""" | |
| # download answers of different models that we trust | |
| download_dataset("Vikhrmodels/openbench-eval", EVAL_RESULTS_PATH) | |
| # print(subprocess.Popen('ls src')) | |
| # copy the trusted model answers to data | |
| subprocess.run( | |
| [ | |
| "rsync", | |
| "-azP", | |
| "--ignore-existing", | |
| f"{EVAL_RESULTS_PATH}/internal/*", | |
| "data/arena-hard-v0.1/model_answer/internal/", | |
| ], | |
| check=False, | |
| ) | |
| # copy the judgement pre generated | |
| # Will be rewritten after we switch to new gen for each submit | |
| subprocess.run( | |
| [ | |
| "rsync", | |
| "-azP", | |
| "--ignore-existing", | |
| f"{EVAL_RESULTS_PATH}/model_judgment/*", | |
| "data/arena-hard-v0.1/model_judgement/", | |
| ], | |
| check=False, | |
| ) | |
| # Retrieve the leaderboard DataFrame | |
| with open("eval-results/evals/upd.json", "r", encoding="utf-8") as eval_file: | |
| leaderboard_df = pd.DataFrame.from_records(json.load(eval_file)) | |
| return leaderboard_df.copy() | |