model-hub

On-demand error localization for a single dataset cell.

Use case: in the dataset detail drawer, the user opens an eval cell that doesn't have an error_analysis block (because the eval was run before error_localization was enabled, or the user wants a fresh run). They click "Run error localization" and we:

  1. Look up the cell + its UserEvalMetric (column.source_id) + the EvalTemplate.
  2. Resolve the metric's mapping (template_var → column UUID) against the row's other cells to build input_data.
  3. Pull the eval verdict + reason from cell.value / cell.value_infos.
  4. Upsert an ErrorLocalizerTask(source=DATASET, source_id=cell.id, status=PENDING) so the existing 30s Temporal schedule picks it up and processes it via process_single_error_localization.

Returns the task id + status. The frontend then polls the cell detail / task status endpoint until error_analysis lands in cell.value_infos.

POST /model-hub/cells/{cell_id}/run-error-localizer/

post/model-hub/cells/{cell_id}/run-error-localizer/

Request

  • Base URL: http://localhost:8000
  • URL: http://localhost:8000/model-hub/cells/{cell_id}/run-error-localizer/
  • Auth: one of:
    • API key in header X-Api-Key
    • API key in header X-Secret-Key

Request body

ModelHubEmptyRequest required

Response

OK

statusboolean required

Changes

No recorded changes to this operation across all 1 revision of this API.