Hugging Face Model Hub

The Hub brings together models, datasets, demos, documentation, versions and permissions.

#Main resources

Use task/framework/language/library filters to find models. Read their model cards for training information, limitations, usage and licensing rather than choosing by download counts alone.

Datasets can offer browser previews and streaming access where supported. Spaces host demonstrations using supported application frameworks and compute tiers; verify resource availability and billing before deployment.

#Download through the SDK

bashCopy
pip install huggingface_hub

Use the installed CLI's supported login flow if access is required.

pythonCopy
from huggingface_hub import snapshot_download, hf_hub_download

snapshot_download(
    repo_id="BAAI/bge-small-zh-v1.5",
    local_dir="./my_local_model",
    revision="main",
)
file_path = hf_hub_download(repo_id="gpt2", filename="config.json")

For reproducible use, prefer a reviewed immutable revision rather than assuming main remains unchanged.

#Uploading is a separate action

pythonCopy
from huggingface_hub import HfApi
api = HfApi()
# These publish repository data; review visibility and contents first.
# api.create_repo(repo_id="username/my-model", exist_ok=True)
# api.upload_folder(folder_path="./model_files",
#                   repo_id="username/my-model", repo_type="model")

#Versions and large files

Hub repositories retain a Git-style versioning model while large-file transport/storage tooling evolves. Use the current supported Git/SDK workflow and record the version used.

Organizations support shared administration. Gated resources can require approval and license acceptance before download. Check access, permitted use and task/language suitability before adopting a model.

Manage the configured cache location and disk needs deliberately rather than deleting unknown cached artifacts.

#References

• Hub documentation↗
• Spaces overview↗