Spaces:
Running
Running
| import os | |
| import pandas as pd | |
| # from src.display.formatting import has_no_nan_values, make_clickable_model | |
| # from src.display.utils import AutoEvalColumn, EvalQueueColumn | |
| # from src.leaderboard.read_evals import get_raw_eval_results | |
| def get_leaderboard_df_crm( | |
| crm_results_path: str, accuracy_cols: list, cost_cols: list | |
| ) -> tuple[pd.DataFrame, pd.DataFrame]: | |
| """Creates a dataframe from all the individual experiment results""" | |
| sf_finetuned_models = ["SF-TextBase 70B", "SF-TextBase 7B", "SF-TextSum"] | |
| leaderboard_accuracy_df = pd.read_csv(os.path.join(crm_results_path, "hf_leaderboard_accuracy.csv")) | |
| leaderboard_accuracy_df = leaderboard_accuracy_df[~leaderboard_accuracy_df["Model Name"].isin(sf_finetuned_models)] | |
| # leaderboard_accuracy_df = leaderboard_accuracy_df.sort_values( | |
| # by=[AutoEvalColumn.accuracy_metric_average.name], ascending=False | |
| # ) | |
| # print(leaderboard_accuracy_df) | |
| # print(leaderboard_accuracy_df.columns) | |
| # print(leaderboard_accuracy_df["Model Name"].nunique()) | |
| leaderboard_accuracy_df = leaderboard_accuracy_df[accuracy_cols].round(decimals=2) | |
| ref_df = leaderboard_accuracy_df[["Model Name", "LLM Provider"]].drop_duplicates() | |
| leaderboard_cost_df = pd.read_csv(os.path.join(crm_results_path, "hf_leaderboard_latency_cost.csv")) | |
| leaderboard_cost_df = leaderboard_cost_df[~leaderboard_cost_df["Model Name"].isin(sf_finetuned_models)] | |
| leaderboard_cost_df = leaderboard_cost_df.join(ref_df.set_index("Model Name"), on="Model Name") | |
| leaderboard_cost_df["LLM Provider"] = leaderboard_cost_df["LLM Provider"].fillna("Google") | |
| leaderboard_cost_df = leaderboard_cost_df[cost_cols].round(decimals=2) | |
| return leaderboard_accuracy_df, leaderboard_cost_df | |
| # 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""" | |
| # raw_data = get_raw_eval_results(results_path, requests_path) | |
| # all_data_json = [v.to_dict() for v in raw_data] | |
| # df = pd.DataFrame.from_records(all_data_json) | |
| # df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False) | |
| # df = df[cols].round(decimals=2) | |
| # # filter out if any of the benchmarks have not been produced | |
| # df = df[has_no_nan_values(df, benchmark_cols)] | |
| # return raw_data, df | |
| # def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]: | |
| # """Creates the different dataframes for the evaluation queues requestes""" | |
| # 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) | |
| # data[EvalQueueColumn.model.name] = make_clickable_model(data["model"]) | |
| # data[EvalQueueColumn.revision.name] = data.get("revision", "main") | |
| # 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 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] | |