Register or begin uploading a dataset
Register a dataset by reference (source: an external dataset reference passed to the trainer as-is, or a repo://{repo}[@{revision}][/{path}] reference reserved for the versioned dataset store) — the dataset is ready immediately. Omit source to upload instead: the response is pending_upload and carries an upload block of presigned PUT targets; PUT the data, then call commit to seal the immutable manifest.
Authorizations
Gateway auth: send Authorization: Bearer <mka1-api-key>. For multi-user server-side integrations, you can also send X-On-Behalf-Of: <external-user-id>.
Headers
Optional external end-user identifier forwarded by the API gateway.
Body
Register a dataset by reference, or begin an upload by omitting source.
Response
Successful Response
A first-class training dataset: a registered reference or an uploaded, committed object set.
How the data is sourced: an opaque external reference the trainer resolves itself, an uploaded object served to the trainer via a presigned fetch URL minted at provider hand-off, or a repo:// reference reserved for the versioned dataset store.
external, upload, repo Total committed bytes; 0 for references and uncommitted uploads.
pending_upload until an uploaded dataset is committed; deleting while asynchronous object removal is in progress.
pending_upload, ready, deleting Immutable record of an uploaded dataset's objects, written once at commit.
The pinned revision parsed from a repo:// source; null otherwise.
The registered source reference for external and repo datasets; null for uploads.
Presigned upload targets for an upload-mode dataset.