Skip to main content
Create a training job by specifying its type, base model, dataset, LoRA configuration, and hyperparameters. Returns a queued job id immediately.

Endpoint

POST /v1/jobs Send an Authorization: Bearer <api_key> header with every request.

Request body

string
required
Training type: sft, opd, pretrain, distill, qat, grpo, dpo, kto, orpo, simpo, or ppo.
string
required
Hugging Face model identifier or local path for the trainable base.
string
Frozen teacher model identifier. Required when type is opd or distill, ignored otherwise.
object
required
LoRA configuration object.
string
required
Dataset identifier registered in PIN.
object
required
Training hyperparameters. Common fields include epochs, lr. Type-specific fields include distill_alpha, temperature, fp8_recipe, max_tokens, qat_group_size, group_size, and beta.
object
Optional auto-intervention configuration with rules, cooldown, and enabled flag.

Response

string
Id of the created job.
string
Initial status, usually queued.

Example

Response