FWfw-ai-external
gateway.openapi_Gateway

Finalize a training run after the trainer has successfully materialized its run-scoped model id. INTERNAL: api_gateway calls this after proxying Tinker create_model to the trainer.

Changed on
post/v1/accounts/{account_id}/trainingRuns/{training_run_id}:finalizeMaterialization

Request

Path parameters

account_idstring required

The Account Id

training_run_idstring required

The Training Run Id

Request body

modelIdstring required

Trainer-returned model id. Must be run-scoped, e.g. "{run_id}:train:0".

Response

A successful response.

namestring
createTimestring date-time
updateTimestring date-time
createdBystring
baseModelstring

Base model resource name the run fine-tunes (accounts/{a}/models/{m}). Set once at create_model and immutable thereafter.

loraRankinteger

LoRA rank for this run. Set once at create_model.

sourceTrainingSessionIdstring

The TrainingSession this run was first created under (its source session). A run is created under a session but persists across sessions — it can later be loaded under a different session — so this records the originating session for lineage. Set once at create and immutable.

state'STATE_UNSPECIFIED' | 'PENDING_MATERIALIZATION' | 'READY' | 'FAILED' | 'EXPIRED'

Materialization state for the trainer-side model backing this run.

UNSPECIFIED is reserved for legacy rows created before the explicit state machine existed; reconcilers must not sweep it.

  • EXPIRED: Clean lifecycle expiry (owning session went idle / was torn down), not a trainer failure. Distinct from FAILED so callers can tell the two apart.

Changes