Help - finding creation dates for Hugging Face models

Hey there!
I’m doing academic research on popular models and datasets from the Hugging Face Hub, and I’m having trouble finding the creation date of each one using the API.

If anyone knows how to retrieve this information programmatically, I’d really appreciate your help


I have the same problem. The Model Hub Search API only provides a lastModified field, which is useless when trying to find trending models, as some old model repositories are just updating their, which in turn updates the whole repositories lastModified date. Also: a lastModified date only for the model files would be useful within the python search API:

the siblings array only returns file names, but no lastModified, and nobody should scrape file dates:

siblings: [ModelFile(rfilename='.gitattributes'), ModelFile(rfilename=''), ModelFile(rfilename='config.json'), ModelFile(rfilename='pytorch_model.bin'), ModelFile(rfilename='sentencepiece.bpe.model'), ModelFile(rfilename='special_tokens_map.json'), ModelFile(rfilename='tokenizer.json'), ModelFile(rfilename='tokenizer_config.json')]


I’ve built a small gradio UI for jupyter notebooks or google colabs to at least get some popular transformer-models and calculate a downloads/days metric:

import gradio as gr
from huggingface_hub import HfApi, ModelFilter
from datetime import datetime
import pandas as pd

# Define a global variable to store the output data
output_data = None
def process_output(data):
    global output_data
    output_data = data

def get_models(task, model_library, date_cutoff, text_search, author):
    api = HfApi()
    models = api.list_models(

    # Convert date_cutoff to datetime object
    date_cutoff = datetime.strptime(date_cutoff, '%Y-%m-%d')

    # Filter models by date_cutoff and calculate downloads per day
    model_data = []
    for model in models:
        last_modified = datetime.strptime(model.lastModified, '%Y-%m-%dT%H:%M:%S.%fZ')
        if last_modified >= date_cutoff:
            days_online = ( - last_modified).days
            downloads_per_day = model.downloads if days_online == 0 else model.downloads / days_online
            model_data.append([model.modelId, last_modified, model.downloads, downloads_per_day])

    # Create a DataFrame and sort by lastModified date
    df = pd.DataFrame(model_data, columns=['Model', 'lastModified', 'totalDownloads', 'downloadsPerDay'])
    df.sort_values(by='lastModified', ascending=False, inplace=True)

    return df

iface = gr.Interface(
        gr.inputs.Dropdown(choices=['text-generation'], label='Task'),
        gr.inputs.Dropdown(choices=['transformers'], label='Libraries'),
        gr.inputs.Textbox(lines=1, label='Date Cutoff (YYYY-MM-DD)'),
        gr.inputs.Textbox(lines=1, label='Text Search'),
        gr.inputs.Textbox(lines=1, label='Author')
    title='Hugging Face Model Explorer',
    description='Explore Hugging Face models by task, architecture, and date.'

#iface.launch(debug=True, share=True)

1 Like


I did some academic research with HFH too, and I noticed this trouble.

Even though there isn’t any solution to this particular concern, our research group published HFCommunity, a relational database with the data of HFH and Git. Then, as we have the commit history, we could consider the first commit timestamp as the creation date of the repo (notice how this isn’t always the case, but to muddle through it is a good starting point).

The database is built with MariaDB and it is offered as a SQL dump. I have retrieved some information about the creation date of some projects.

Don’t hesitate to ask any doubt regarding the use or extraction of HFC data.

PD: Any feedback would be greatly appreciated

1 Like