Spaces:
Sleeping
Sleeping
import json | |
import gradio as gr | |
import pandas as pd | |
from huggingface_hub import HfFileSystem | |
from src.constants import RESULTS_DATASET_ID, TASKS | |
def fetch_result_paths(): | |
fs = HfFileSystem() | |
paths = fs.glob(f"{RESULTS_DATASET_ID}/**/**/*.json") | |
return paths | |
def filter_latest_result_path_per_model(paths): | |
from collections import defaultdict | |
d = defaultdict(list) | |
for path in paths: | |
model_id, _ = path[len(RESULTS_DATASET_ID) + 1:].rsplit("/", 1) | |
d[model_id].append(path) | |
return {model_id: max(paths) for model_id, paths in d.items()} | |
def update_load_results_component(): | |
return gr.Button("Load Results", interactive=True) | |
def load_results_dataframe(model_id, result_path_per_model=None): | |
if not model_id or not result_path_per_model: | |
return | |
result_path = result_path_per_model[model_id] | |
fs = HfFileSystem() | |
with fs.open(result_path, "r") as f: | |
data = json.load(f) | |
model_name = data.get("model_name", "Model") | |
df = pd.json_normalize([{key: value for key, value in data.items()}]) | |
# df.columns = df.columns.str.split(".") # .split return a list instead of a tuple | |
return df.set_index(pd.Index([model_name])).reset_index() | |
def load_results_dataframes(*model_ids, result_path_per_model=None): | |
return [load_results_dataframe(model_id, result_path_per_model=result_path_per_model) for model_id in model_ids] | |
def display_results(task, *dfs): | |
dfs = [df.set_index("index") for df in dfs if "index" in df.columns] | |
if not dfs: | |
return None, None | |
df = pd.concat(dfs) | |
df = df.T.rename_axis(columns=None) | |
return display_tab("results", df, task), display_tab("configs", df, task) | |
def display_tab(tab, df, task): | |
df = df.style.format(na_rep="") | |
df.hide( | |
[ | |
row | |
for row in df.index | |
if ( | |
not row.startswith(f"{tab}.") | |
or row.startswith(f"{tab}.leaderboard.") | |
or row.endswith(".alias") | |
or (not row.startswith(f"{tab}.{task}") if task != "All" else False) | |
) | |
], | |
axis="index", | |
) | |
start = len(f"{tab}.leaderboard_") if task == "All" else len(f"{tab}.{task} ") | |
df.format_index(lambda idx: idx[start:].removesuffix(",none"), axis="index") | |
return df.to_html() | |
def update_tasks_component(): | |
return gr.Radio( | |
["All"] + list(TASKS.values()), | |
label="Tasks", | |
info="Evaluation tasks to be displayed", | |
value="All", | |
interactive=True, | |
) | |
def clear_results(): | |
# model_id_1, model_id_2, dataframe_1, dataframe_2, task | |
return ( | |
None, None, None, None, | |
gr.Radio( | |
["All"] + list(TASKS.values()), | |
label="Tasks", | |
info="Evaluation tasks to be displayed", | |
value="All", | |
interactive=False, | |
), | |
) | |