app.py

from huggingface_hub import list_models
from cachetools import cached, TTLCache
from toolz import groupby, valmap
import gradio as gr
from tqdm.auto import tqdm
import pandas as pd


@cached(TTLCache(maxsize=10, ttl=60 * 60 * 3))
def get_all_models():
    models = list(tqdm(iter(list_models(cardData=True))))
    return [model for model in models if model is not None]


def has_base_model_info(model):
    try:
        if card_data := model.cardData:
            if base_model := card_data.get("base_model"):
                if isinstance(base_model, str):
                    return True
    except AttributeError:
        return False
    return False


grouped_by_has_base_model_info = groupby(has_base_model_info, get_all_models())
print(valmap(len, grouped_by_has_base_model_info))

summary = f"""{len(grouped_by_has_base_model_info.get(True))} models have base model info. 
            {len(grouped_by_has_base_model_info.get(False))} models don't have base model info.
            Currently {round(len(grouped_by_has_base_model_info.get(True))/len(get_all_models())*100,2)}% of models have base model info."""

models_with_base_model_info = grouped_by_has_base_model_info.get(True)
base_models = [
    model.cardData.get("base_model") for model in models_with_base_model_info
]
df = pd.DataFrame(
    pd.DataFrame({"base_model": base_models}).value_counts()
).reset_index()

grouped_by_base_model = groupby(
    lambda x: x.cardData.get("base_model"), models_with_base_model_info
)

all_base_models = df["base_model"].to_list()


def return_models_for_base_model(base_model):
    models = grouped_by_base_model.get(base_model)
    # sort models by downloads
    models = sorted(models, key=lambda x: x.downloads, reverse=True)
    results = ""
    results += f"## {base_model} children\n\n"
    results += f"{base_model} has {len(models)} children\n\n"
    for model in models:
        url = f"https://huggingface.co/{model.modelId}"
        results += (
            f"[{model.modelId}]({url}) | number of downloads {model.downloads}" + "\n\n"
        )
    return results


with gr.Blocks() as demo:
    gr.Markdown("### Models with base model info")
    gr.Markdown(summary)
    gr.Markdown("### Find all models trained from a base model")
    base_model = gr.Dropdown(all_base_models, label="Base Model")
    results = gr.Markdown()
    base_model.change(return_models_for_base_model, base_model, results)

    # gr.DataFrame(df)
demo.launch()