logikon/open_cot_leaderboard

1import json2import os3 4import pandas as pd5 6from src.display.formatting import has_no_nan_values, make_clickable_model7from src.display.utils import AutoEvalColumn, EvalQueueColumn8from src.leaderboard.read_evals import get_raw_eval_results9 10 11def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:12    raw_data = get_raw_eval_results(results_path, requests_path)13    all_data_json = [v.to_dict() for v in raw_data]14 15    df = pd.DataFrame.from_records(all_data_json)16    df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)17    df = df[cols].round(decimals=2)18 19    # filter out if any of the benchmarks have not been produced20    df = df[has_no_nan_values(df, benchmark_cols)]21    return raw_data, df22 23 24def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:25    entries = [entry for entry in os.listdir(save_path) if not entry.startswith(".")]26    all_evals = []27 28    for entry in entries:29        if ".json" in entry:30            file_path = os.path.join(save_path, entry)31            with open(file_path) as fp:32                data = json.load(fp)33 34            data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])35            data[EvalQueueColumn.revision.name] = data.get("revision", "main")36 37            all_evals.append(data)38        elif ".md" not in entry:39            # this is a folder40            sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if not e.startswith(".")]41            for sub_entry in sub_entries:42                file_path = os.path.join(save_path, entry, sub_entry)43                with open(file_path) as fp:44                    data = json.load(fp)45 46                data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])47                data[EvalQueueColumn.revision.name] = data.get("revision", "main")48                all_evals.append(data)49 50    pending_list = [e for e in all_evals if e["status"] in ["PENDING", "RERUN"]]51    pending_list = sorted(pending_list, key=lambda x: x.get("date","ZZZ"))52    running_list = [e for e in all_evals if e["status"] == "RUNNING"]53    finished_list = [e for e in all_evals if e["status"].startswith("FINISHED") or e["status"] == "PENDING_NEW_EVAL"]54    df_pending = pd.DataFrame.from_records(pending_list, columns=cols)55    df_running = pd.DataFrame.from_records(running_list, columns=cols)56    df_finished = pd.DataFrame.from_records(finished_list, columns=cols)57    return df_finished[cols], df_running[cols], df_pending[cols]58