Code beginner · 3 min read

How to use Hugging Face hub in python

Q: How to use Hugging Face hub in python

Use the huggingface_hub Python library to authenticate, download, and interact with models from the Hugging Face Hub programmatically.

Direct answer

Use the huggingface_hub Python library to authenticate, download, and interact with models from the Hugging Face Hub programmatically.

Setup

Install

bash

pip install huggingface_hub

Env vars

HF_API_TOKEN

Imports

python

from huggingface_hub import HfApi, hf_hub_download
import os

Examples

inDownload the 'bert-base-uncased' tokenizer files

outFiles downloaded locally with their paths returned

inList models under the 'facebook' organization

outList of model names like ['facebook/bart-large', 'facebook/mbart-large-50']

inDownload a specific file from a private repo using token

outFile downloaded successfully with access token authentication

Integration steps

Install the huggingface_hub package via pip
Set your Hugging Face API token in the HF_API_TOKEN environment variable
Import HfApi and hf_hub_download from huggingface_hub
Initialize HfApi client with the token from environment
Use HfApi methods to list or get model info
Use hf_hub_download to download model files locally

Full code

python

from huggingface_hub import HfApi, hf_hub_download
import os

# Load your Hugging Face API token from environment
api_token = os.environ.get('HF_API_TOKEN')

# Initialize the API client
api = HfApi(token=api_token)

# Example 1: List models by an organization
models = api.list_models(author='facebook', limit=5)
print('Facebook models:')
for model in models:
    print(f'- {model.modelId}')

# Example 2: Download a file from a model repo
model_id = 'bert-base-uncased'
filename = 'config.json'
file_path = hf_hub_download(repo_id=model_id, filename=filename, token=api_token)
print(f'Downloaded {filename} to {file_path}')

output

Facebook models:
- facebook/bart-large
- facebook/mbart-large-50
- facebook/wav2vec2-base-960h
- facebook/detr-resnet-50
- facebook/bart-large-mnli
Downloaded config.json to /home/user/.cache/huggingface/hub/models--bert-base-uncased/config.json

API trace

Request

json

{"author": "facebook", "limit": 5, "token": "<your_token>"} for list_models; {"repo_id": "bert-base-uncased", "filename": "config.json", "token": "<your_token>"} for hf_hub_download

Response

json

{"models": [{"modelId": "facebook/bart-large"}, ...]} for list_models; local file path string for hf_hub_download

ExtractFor list_models: iterate response.models and access modelId; for hf_hub_download: use returned file path string

Variants

Streaming download with progress bar ›

Use when downloading large files to support resuming and show progress.

python

from huggingface_hub import hf_hub_download
import os

api_token = os.environ.get('HF_API_TOKEN')

file_path = hf_hub_download(
    repo_id='bert-base-uncased',
    filename='pytorch_model.bin',
    token=api_token,
    resume_download=True
)
print(f'Downloaded model weights to {file_path}')

Anonymous access without token ›

Use for public models where authentication is not required.

python

from huggingface_hub import hf_hub_download

file_path = hf_hub_download(repo_id='bert-base-uncased', filename='config.json')
print(f'Downloaded {file_path}')

Using HfApi to upload a model file ›

Use when you want to programmatically upload files to your Hugging Face model repository.

python

from huggingface_hub import HfApi
import os

api_token = os.environ.get('HF_API_TOKEN')
api = HfApi(token=api_token)

api.upload_file(
    path_or_fileobj='local_model.bin',
    path_in_repo='model.bin',
    repo_id='your-username/your-model',
    repo_type='model'
)
print('File uploaded successfully')

Performance

Latency~200-500ms for metadata requests; file downloads depend on file size and network speed

CostFree for public models; private repo access requires a valid token but no additional cost

Rate limitsHugging Face enforces rate limits per user/token; typically 60 requests per minute for API calls

Cache downloaded files locally to avoid repeated downloads
Use specific filenames to minimize data transfer
Limit list_models queries with 'limit' parameter to reduce overhead

Approach	Latency	Cost/call	Best for
HfApi list_models	~200ms	Free	Listing models and metadata
hf_hub_download	Varies by file size	Free	Downloading model files
Anonymous access	~200ms	Free	Public models without authentication

✓

Quick tip

Always store your Hugging Face API token in an environment variable and pass it to the client to access private models securely.

⚠

Common mistake

Forgetting to set or pass the HF_API_TOKEN environment variable causes authentication errors when accessing private repos.

Verified 2026-04

Verify ↗