Spaces:
Running
Running
| import json | |
| import os | |
| import pandas as pd | |
| from src.display.formatting import has_no_nan_values, make_clickable_model | |
| from src.display.utils import AutoEvalColumn, AutoEvalColumnMultimodal, EvalQueueColumn | |
| from src.leaderboard.read_evals import get_raw_eval_results, get_raw_eval_results_mib | |
| def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame: | |
| """Creates a dataframe from all the individual experiment results""" | |
| print(f"results_path is {results_path}, requests_path is {requests_path}") | |
| raw_data = get_raw_eval_results(results_path, requests_path) | |
| print(f"raw_data is {raw_data}") | |
| all_data_json = [v.to_dict() for v in raw_data] | |
| print(f"all_data_json is {pd.DataFrame.from_records(all_data_json)}") | |
| all_data_json_filtered = [] | |
| for item in all_data_json: | |
| item["Track"] = item["eval_name"].split("_")[-1] | |
| item["ioi"] = 0 | |
| item["mcqa"] = 0 | |
| if "VQA" in benchmark_cols and "VQA" in item: | |
| all_data_json_filtered.append(item) | |
| if "VQA" not in benchmark_cols and "VQA" not in item: | |
| all_data_json_filtered.append(item) | |
| all_data_json = all_data_json_filtered | |
| df = pd.DataFrame.from_records(all_data_json) | |
| df = df.sort_values(by=[AutoEvalColumn.text_average.name], ascending=False) | |
| # df = df.sort_values(by=[Tasks.task0.value.col_name], ascending=False) | |
| # df = df.sort_values(by=[AutoEvalColumn.track.name], ascending=False) | |
| print(f"df is {df}") | |
| # df = df[cols].round(decimals=1) | |
| # filter out if any of the benchmarks have not been produced | |
| df = df[has_no_nan_values(df, benchmark_cols)] | |
| return df | |
| # def get_leaderboard_df_mib(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame: | |
| # """Creates a dataframe from all the individual experiment results""" | |
| # print(f"results_path is {results_path}, requests_path is {requests_path}") | |
| # raw_data = get_raw_eval_results(results_path, requests_path) | |
| # print(f"raw_data is {raw_data}") | |
| # all_data_json = [v.to_dict() for v in raw_data] | |
| # print(f"all_data_json is {pd.DataFrame.from_records(all_data_json)}") | |
| # all_data_json_filtered = [] | |
| # for item in all_data_json: | |
| # item["Track"] = item["eval_name"].split("_")[-1] | |
| # if "VQA" in benchmark_cols and "VQA" in item: | |
| # all_data_json_filtered.append(item) | |
| # if "VQA" not in benchmark_cols and "VQA" not in item: | |
| # all_data_json_filtered.append(item) | |
| # all_data_json_filtered.append(item) | |
| # all_data_json = all_data_json_filtered | |
| # df = pd.DataFrame.from_records(all_data_json) | |
| # df = df.sort_values(by=[AutoEvalColumn.text_average.name], ascending=False) | |
| # print(f"df is {df}") | |
| # df = df[cols].round(decimals=1) | |
| # # filter out if any of the benchmarks have not been produced | |
| # df = df[has_no_nan_values(df, benchmark_cols)] | |
| # return df | |
| def get_leaderboard_df_mib(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame: | |
| """Creates a dataframe from all the MIB experiment results""" | |
| print(f"results_path is {results_path}, requests_path is {requests_path}") | |
| raw_data = get_raw_eval_results_mib(results_path, requests_path) | |
| print(f"raw_data is {raw_data}") | |
| # Convert each result to dict format | |
| all_data_json = [v.to_dict() for v in raw_data] | |
| print(f"all_data_json is {pd.DataFrame.from_records(all_data_json)}") | |
| # Convert to dataframe | |
| df = pd.DataFrame.from_records(all_data_json) | |
| # Sort by Average score descending | |
| if 'Average' in df.columns: | |
| # Convert '-' to NaN for sorting purposes | |
| df['Average'] = pd.to_numeric(df['Average'], errors='coerce') | |
| df = df.sort_values(by=['Average'], ascending=False, na_position='last') | |
| # Convert NaN back to '-' | |
| df['Average'] = df['Average'].fillna('-') | |
| return df | |
| def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]: | |
| """Creates the different dataframes for the evaluation queues requests""" | |
| entries = [entry for entry in os.listdir(save_path) if not entry.startswith(".")] | |
| all_evals = [] | |
| for entry in entries: | |
| if ".json" in entry: | |
| file_path = os.path.join(save_path, entry) | |
| with open(file_path) as fp: | |
| data = json.load(fp) | |
| if "still_on_hub" in data and data["still_on_hub"]: | |
| data[EvalQueueColumn.model.name] = make_clickable_model(data["hf_repo"], data["model"]) | |
| data[EvalQueueColumn.revision.name] = data.get("revision", "main") | |
| else: | |
| data[EvalQueueColumn.model.name] = data["model"] | |
| data[EvalQueueColumn.revision.name] = "N/A" | |
| all_evals.append(data) | |
| elif ".md" not in entry: | |
| # this is a folder | |
| sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if os.path.isfile(e) and not e.startswith(".")] | |
| for sub_entry in sub_entries: | |
| file_path = os.path.join(save_path, entry, sub_entry) | |
| with open(file_path) as fp: | |
| data = json.load(fp) | |
| data[EvalQueueColumn.model.name] = make_clickable_model(data["model"]) | |
| data[EvalQueueColumn.revision.name] = data.get("revision", "main") | |
| all_evals.append(data) | |
| pending_list = [e for e in all_evals if e["status"] in ["PENDING", "RERUN"]] | |
| running_list = [e for e in all_evals if e["status"] == "RUNNING"] | |
| finished_list = [e for e in all_evals if e["status"].startswith("FINISHED") or e["status"] == "PENDING_NEW_EVAL"] | |
| df_pending = pd.DataFrame.from_records(pending_list, columns=cols) | |
| df_running = pd.DataFrame.from_records(running_list, columns=cols) | |
| df_finished = pd.DataFrame.from_records(finished_list, columns=cols) | |
| return df_finished[cols], df_running[cols], df_pending[cols] |