Skip to content
Docs
Hub and Collaboration
Model repositories

Model Repositories

🚧
Tensormorph does not have a confirmed, branded hosting product of its own - there is no "Tensormorph Hub" documented in its design as of this writing. This page covers where models Tensormorph opens actually live today, and the nearest confirmed future direction.

Where models come from today

Tensormorph is a consumer of checkpoints, not a host for them. Two confirmed sources:

  • Local files - any supported format on disk; see Working with Models.
  • Hugging Face Hub - Tensormorph can open a model directly by its Hub reference and revision, using Hugging Face's own repository and caching model; see Hugging Face Models. This is a repository system Tensormorph reads from, not one it defines or hosts.

Weight history, separately, lives in Lob - an external system Tensormorph anchors to (see Versions and Branches) rather than a repository product Tensormorph itself provides.

The nearest confirmed future direction

A Tile Server with a thin web client is a confirmed, lower-priority item on Tensormorph's roadmap - a way to serve a checkpoint's tiles and statistics over a network for lightweight remote viewing, closer to a serving layer than to a full repository-hosting product with its own accounts, search, and discovery UI. Don't assume it implies anything beyond that.

If you're looking for repository features today

For hosting, discovery, access control, and sharing a checkpoint with others before Tensormorph has its own answer to this, the Hugging Face Hub (or an equivalent your organization already uses) remains the practical option - Tensormorph opening a Hub reference directly means you don't lose inspection capability by keeping models there in the meantime.

Related resources

  • Hugging Face Models - the confirmed way Tensormorph reads Hub-hosted checkpoints today.
  • Versions and Branches - where weight history actually lives (Lob), distinct from repository hosting.
  • Permissions - access control, which is similarly not yet a Tensormorph-native feature.