Spaces:
Running
Running
import fire | |
import json | |
import pandas as pd | |
import pickle | |
def main( | |
model_info_file: str, | |
elo_rating_pkl: str, | |
output_csv: str | |
): | |
model_info = json.load(open(model_info_file)) | |
with open(elo_rating_pkl, "rb") as fin: | |
elo_rating_results = pickle.load(fin) | |
# Model, Dim Elo rating (anony), Arena Elo rating (anony), Link, Orgnization | |
model_ratings = model_info | |
fields = ["key", "Model"] | |
for dim, dim_results in elo_rating_results.items(): | |
anony_elo_rating_results = dim_results["anony"] | |
full_elo_rating_results = dim_results["full"] | |
anony_leaderboard_data = anony_elo_rating_results["leaderboard_table_df"] | |
full_leaderboard_data = full_elo_rating_results["leaderboard_table_df"] | |
fields += [f"{dim} Elo rating"] | |
all_models = anony_leaderboard_data.index.tolist() | |
for model in all_models: | |
if not model in model_ratings: | |
# set Organization and license to empty | |
model_ratings[model] = {} | |
model_ratings[model]["Organization"] = "N/A" | |
model_ratings[model]["Link"] = "N/A" | |
model_ratings[model]["Model"] = model | |
model_ratings[model]["key"] = model | |
if model in anony_leaderboard_data.index: | |
model_ratings[model][f"{dim} Elo rating"] = anony_leaderboard_data.loc[model, "rating"] | |
else: | |
model_ratings[model][f"{dim} Elo rating"] = 0 | |
if "Arena Elo rating" not in model_ratings[model].keys(): | |
model_ratings[model]["Arena Elo rating"] = 0 | |
model_ratings[model]["Arena Elo rating"] += model_ratings[model][f"{dim} Elo rating"] | |
## Anony | |
# if model in anony_leaderboard_data.index: | |
# model_ratings[model][f"{dim} Elo rating (anony)"] = anony_leaderboard_data.loc[model, "rating"] | |
# else: | |
# model_ratings[model][f"{dim} Elo rating (anony)"] = 0 | |
# if "Arena Elo rating (anony)" not in model_ratings[model].keys(): | |
# model_ratings[model]["Arena Elo rating (anony)"] = 0 | |
# model_ratings[model]["Arena Elo rating (anony)"] += model_ratings[model][f"{dim} Elo rating (anony)"] | |
## Anony + Named | |
# if model in full_elo_rating_results["leaderboard_table_df"].index: | |
# model_ratings[model][f"{dim} Elo rating (full)"] = full_leaderboard_data.loc[model, "rating"] | |
# else: | |
# model_ratings[model][f"{dim} Elo rating (full)"] = 0 | |
# if "Arena Elo rating (full)" not in model_ratings[model].keys(): | |
# model_ratings[model]["Arena Elo rating (full)"] = 0 | |
# model_ratings[model]["Arena Elo rating (full)"] += model_ratings[model][f"{dim} Elo rating (full)"] | |
fields += ["Arena Elo rating", "Link", "Organization"] | |
# fields += ["Arena Elo rating (anony)", "Arena Elo rating (full)", "Link", "Organization"] | |
final_model_info = {} | |
print(model_ratings) | |
for model in model_ratings: | |
if "Model" in model_ratings[model]: | |
# model_ratings[model]["Arena Elo rating (anony)"] /= 5 | |
# model_ratings[model]["Arena Elo rating (full)"] /= 5 | |
model_ratings[model]["Arena Elo rating"] /= 5 | |
final_model_info[model] = model_ratings[model] | |
model_info = final_model_info | |
exclude_keys = ['starting_from'] | |
for key in exclude_keys: | |
for model in model_info: | |
if key in model_info[model]: | |
del model_info[model][key] | |
df = pd.DataFrame(model_info).T | |
df = df[fields] | |
# sort by anony rating | |
df = df.sort_values(by=["Arena Elo rating"], ascending=False) | |
df.to_csv(output_csv, index=False) | |
print("Leaderboard data saved to", output_csv) | |
print(df) | |
if __name__ == "__main__": | |
fire.Fire(main) |