Update dataset
Update a dataset.
Only the fields in the request change; omitted fields keep their value. Setting a schema or transformations checks every existing example against the result. Returns 400, leaving the dataset unchanged, if an example does not match, the dataset is not a key-value dataset, or a transformation points into inputs or outputs that have no schema. The X-Updated-Examples-Count response header holds the number of examples changed through patch_examples. Returns 404 if the dataset does not exist and 409 if the new name is already used.
Authorizations
Path Parameters
Body
Update class for Dataset.
New dataset name. Must be unique in the workspace; a name already used by another dataset returns 409. Null returns 400. Omit to keep the current name.
New dataset description. Null clears it. Omit to keep the current description.
JSON Schema that example inputs must match. Replaces the current inputs schema as a whole; null removes it, which returns 400 while transformations still point into inputs. Only key-value datasets support schemas. Every existing example must match the new schema, otherwise the request returns 400 with the IDs of the examples that do not, and nothing is changed. Omit to keep the current schema.
- Inputs Schema Definition
- Missing
JSON Schema that example outputs must match. Replaces the current outputs schema as a whole; null removes it, which returns 400 while transformations still point into outputs. Only key-value datasets support schemas. Every existing example must match the new schema, otherwise the request returns 400 with the IDs of the examples that do not, and nothing is changed. Omit to keep the current schema.
- Outputs Schema Definition
- Missing
Example updates keyed by example ID, applied in the same transaction so existing examples can be fixed to match a new schema. Only used when the request also sets inputs_schema_definition, outputs_schema_definition, or transformations and the dataset ends up with at least one schema; otherwise it is ignored. Each update replaces the inputs, outputs, metadata, or split it sets and keeps the rest; attachment operations are ignored. An ID that is not an example in this dataset returns 404.
- DatasetTransformation · object[]
- Missing · object
Dataset metadata as a JSON object. Replaces the current metadata as a whole; keys are not merged. Send an empty object to clear it. Omit to keep the current metadata.
- Metadata
- Missing
ID of the experiment used as the baseline when comparing experiments on this dataset. Must be an experiment run on this dataset, otherwise returns 400. Null clears it. Omit to keep the current baseline.
Response
Dataset updated successfully
Enum for dataset data types.
kv, llm, chat 
