Jan 5, 2024
Core APIs
Start Training a Model
Kick off fine‑tuning on your datastore with a single training_quality setting and optional evaluation dataset.
Endpoint
Payload:
training_quality (string, required): One of:
low(fast, lightweight tuning, fewer epochs)medium(balanced tuning)high(longer tuning for best accuracy)
evaluation_data_uri (string, optional): URI to a CSV/Parquet containing hold‑out interactions for evaluation; must conform to Interactions schema.
The system determines model base, batch size, learning rate, epochs, and GPU allocation based solely on training_quality internally. You do not need to specify low‑level hyperparameters.
Response (202 Accepted):
Field | Type | Description |
train_job_id | string | Identifier to poll job status. |
status | string |
|
submitted_at | string | UTC timestamp when job was accepted. |
training_quality | string | Echo of requested quality ( |