Custom Training

Page Outline

Path Parameters
Request Body
Behavior
Response

Custom Models Only

This endpoint is only applicable to models with model_type: "custom". It is not available for AutoML (model_type: "auto") models.

Triggers a custom training pipeline for a model, executing the user-supplied training script against the linked dataset within a sandboxed execution environment.

Base Path: /api/v1/models

Method: POST

Path: /{model_id}/custom/train

Auth: Requires a valid Bearer token (Authorization: Bearer <JWT>)

Path Parameters

Parameter	Type	Description
`model_id`	string	Unique identifier of the custom model

Request Body

{
  "entry_point": "train.py",
  "runtime": "python3.11",
  "environment_vars": {
    "LEARNING_RATE": "0.001",
    "EPOCHS": "50"
  },
  "compute": {
    "cpu_cores": 4,
    "memory_gb": 16,
    "gpu": false
  },
  "run_label": "custom-run-experiment-1",
  "timeout_seconds": 14400
}

Field	Type	Required	Description
`entry_point`	string	✅	Filename of the uploaded script asset to execute (must match an uploaded `asset_type: "script"` asset)
`runtime`	string	✅	Execution runtime: `python3.10`, `python3.11`, `python3.12`
`environment_vars`	object	❌	Key-value environment variables passed to the training script
`compute.cpu_cores`	integer	❌	Number of CPU cores to allocate (default: 2, max: 32)
`compute.memory_gb`	integer	❌	RAM in GB to allocate (default: 8, max: 128)
`compute.gpu`	boolean	❌	If `true`, a GPU worker is requested (subject to availability)
`run_label`	string	❌	Human-readable label for this training run
`timeout_seconds`	integer	❌	Max execution time before auto-cancellation (default: 7200, max: 86400)

Behavior

Requires at least one uploaded asset_type: "script" asset. Returns HTTP 422 if none exist.
The training script operates within a sandboxed container. It receives the linked dataset path via the DATASET_PATH environment variable.
The script is expected to write the serialized model artifact to the path specified in the MODEL_OUTPUT_PATH environment variable; otherwise no version is created upon completion.
Model status transitions to training immediately upon a successful submission.
Returns HTTP 409 if a training run is already in progress.
An audit log entry is written with action model.custom_training_started.
Poll progress via the Get Metrics or Get Logs endpoints using the returned run_id.

Response

Returns a CustomTrainingRun object.

{
  "run_id": "run-custom-20260313-f7e2a1",
  "model_id": "mdl-7f3a19bc",
  "model_type": "custom",
  "status": "queued",
  "entry_point": "train.py",
  "runtime": "python3.11",
  "compute": {
    "cpu_cores": 4,
    "memory_gb": 16,
    "gpu": false
  },
  "run_label": "custom-run-experiment-1",
  "created_at": "2026-03-13T16:10:00Z",
  "estimated_start_time": "2026-03-13T16:12:00Z"
}

Path Parameters​

Request Body​

Behavior​

Response​

Path Parameters

Request Body

Behavior

Response