Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import re | |
| import json | |
| from huggingface_hub import HfApi | |
| import pandas as pd | |
| import os | |
| token = os.getenv('HF_API_KEY') | |
| print(f"The token is {token}") | |
| api = HfApi(token=token) | |
| def compute_df(): | |
| # download all files in https://huggingface.co/illuin-cde/baselines | |
| files = [f for f in api.list_repo_files("illuin-cde/baselines-v3") if f.startswith("metrics")] | |
| print(files) | |
| metrics = [] | |
| cols = ["model", "is_contextual", "pooling"] | |
| for file in files: | |
| result_path = api.hf_hub_download("illuin-cde/baselines-v3", filename=file) | |
| with open(result_path, "r") as f: | |
| dic = json.load(f) | |
| if dic["is_contextual"] not in ["multi", "single", "hybrid"]: | |
| continue | |
| # prefix = dic.get("pooling", "default") | |
| dic["model"] = dic.get("model").split("/")[-1] # + "-" + prefix | |
| metrics_cur = dic["metrics"] | |
| for k, v in metrics_cur.items(): | |
| dic.update({k: v["ndcg_at_10"]}) | |
| if k not in cols: | |
| cols.append(k) | |
| del dic["metrics"] | |
| metrics.append(dic) | |
| df = pd.DataFrame(metrics) | |
| df = df[cols] | |
| # df["model"] = df["model"].apply(lambda x: x.split("/")[-1]) | |
| # round all numeric columns | |
| # avg all numeric columns | |
| df["avg"] = df.iloc[:, 3:].mean(axis=1) | |
| df = df.round(3) | |
| # sort by ndcg_at_5 | |
| df = df.sort_values(by="avg", ascending=False) | |
| # gradio display | |
| # gradio_df = gr.Dataframe(df) | |
| # return gradio_df | |
| return df | |
| gradio_df = compute_df() | |
| def refresh(model_query_regex): | |
| if not model_query_regex: | |
| return gradio_df | |
| # filter the dataframe based on regex | |
| return gr.Dataframe(gradio_df[gradio_df["model"].str.contains(model_query_regex, flags=re.IGNORECASE, regex=True)]) | |
| # refresh button and precompute | |
| # gr.Interface(fn=compute_df, title="Results Leaderboard", inputs=None, outputs="dataframe").launch() | |
| gr.Interface(fn=refresh, title="Results Leaderboard", inputs=gr.Textbox(label="Search Model"), outputs="dataframe").launch() |