app.py

from huggingface_hub import list_datasets
from toolz import concat
from toolz import frequencies
import gradio as gr
import pandas as pd

from huggingface_hub import list_models

from huggingface_hub import ModelFilter
from tqdm.auto import tqdm

"""## Grab datasets with `token-classification` task"""

datasets = list(list_datasets(filter="task_categories:token-classification"))

datasets[0]

with_language = [
    dataset
    for dataset in datasets
    if "language" in [t.split(":")[0] for t in dataset.tags]
]

with_language[0]


def get_languages(dataset):
    tags = list(dataset.tags)
    languages = [t for t in tags if t.split(":")[0] == "language"]
    languages = [language for language in languages if len(language.split(":")) == 2]
    return [t.split(":")[1] for t in languages]


freqs = frequencies(concat(get_languages(dataset) for dataset in with_language))
freqs = dict(sorted(freqs.items(), key=lambda x: x[1], reverse=True))


no_model = []
for lang in tqdm(freqs.keys()):
    models_for_lang_with_task_token_classification = list(
        list_models(filter=ModelFilter(language=lang, task="token-classification"))
    )
    models_for_lang_any_task = list(list_models(filter=ModelFilter(language=lang)))
    datasets_for_lang_any_task = list(list_datasets(filter=f"language:{lang}"))
    if not models_for_lang_with_task_token_classification:
        data = {
            "language": lang,
            "datasets_for_token_classification": freqs[lang],
            "datasets": len(datasets_for_lang_any_task),
            "token_classification_models": len(
                models_for_lang_with_task_token_classification
            ),
            "all_models": len(models_for_lang_any_task),
        }
        no_model.append(data)

len(no_model)


df = pd.DataFrame(no_model)

df = df.sort_values(
    by=[
        "datasets_for_token_classification",
        "datasets",
        "token_classification_models",
        "all_models",
    ],
    ascending=[False, False, True, True],
)


def report_summary():
    summary = ""
    for row in df.head(30).itertuples():
        language = row[1]
        summary += f"# Summary for language: {language}\n"
        summary += f"This language has {(row[2])} token classification datasets, {row[3]} datasets overall, {row[4]} token classification models, and {row[5]} models overall.\n"
        summary += f"- [Datasets for token classification task for {language}](https://huggingface.co/datasets?task_categories=task_categories:token-classification&language=language:{language})\n"
        summary += f"- [Token classification models for {language}](https://huggingface.co/models?task_categories=task_categories:token-classification&language=language:{language})\n"
        summary += f"- [All models for {language}](https://huggingface.co/models?language={language}&sort=trending)\n"
        summary += "<br>\n"
    return summary


with gr.Blocks() as demo:
    gr.DataFrame(df)
    gr.Markdown("# Top 30 candidates")
    gr.Markdown(
        "Candiates generated by sorting by most token classification datasets, then least token classification models, then least models overall"
    )
    gr.Markdown(report_summary())


demo.launch()