Skip to main content

Training - TypeScript SDK

client.training manages fine-tuning jobs. Create requests use logical model, Dataset, recipe, and placement references; the API does not accept raw paths, container images, commands, environments, or credentials.

Overview​

Available Operations​

MethodDescription
capabilities()Read the active recipe/resource catalog.
clusterOptions()List clusters selectable for Training.
preview(body)Preflight a resource specification in a cluster.
knowledgeTeacherModels(options)Find compatible knowledge-distillation teachers.
create(body)Create a fine-tuning job.
list(body?)Page Training jobs.
get(id)Read job progress, losses, actions, and artifacts.
artifactDownloadUrl(artifactId)Create temporary download URLs for a completed artifact.
cancel(id)Request cancellation of a non-terminal job.

capabilities​

Read the active recipe/resource catalog.

Request​

This method has no parameters and sends no request body.

Response​

TrainCapabilitiesVO.

clusterOptions​

List clusters selectable for Training.

Request​

This method has no parameters and sends no request body.

Response​

WorkloadClusterOptionVO[].

preview​

Preflight a resource specification in a cluster.

Request​

clusterId, resourceSpecId required.

Response​

WorkloadAdmissionPreviewVO.

knowledgeTeacherModels​

Find compatible knowledge-distillation teachers.

Request​

{studentModelId, recipeId} required.

Response​

TrainKnowledgeTeacherVO[].

create​

Create a fine-tuning job.

Request​

Fields below.

Response​

{jobId}.

list​

Page Training jobs.

Request​

Optional pageNum, pageSize, status.

Response​

PageResult<TrainJobDetailVO>.

get​

Read job progress, losses, actions, and artifacts.

Request​

id: string required.

Response​

TrainJobDetailVO.

artifactDownloadUrl​

Create temporary download URLs for a completed artifact.

Request​

artifactId: string required.

Response​

{urls: string[], files?: Array<{path, sizeBytes, url}>}.

cancel​

Request cancellation of a non-terminal job.

Request​

id: string required.

Response​

null.

Field Reference and Examples​

Create body fields

FieldTypeDescription
clientTokenstringIdempotency token for safe create retries.
displayNamestringHuman-facing job name.
outputModelNamestringOutput model suffix. Its maximum length depends on baseModelRef.id because the final deployable name is <base>-FT-<suffix>; use the capability limit (the tested 4B Profile permits 14 characters).
recipeId, recipeVersionstringRecipe identity from capabilities().
baseModelRefobject{type, id} logical base-model reference.
teacherRefobjectOptional teacher reference; fields described below.
datasetRefsArray<object>Each item uses {datasetId, role}.
placementobject{clusterId, nodeId?, resourceSpecId}.
paramsobjectRecipe hyperparameters.

teacherRef uses snake_case wire keys: provider_key_id, model_id, service_id, and model_asset_id, plus type. Do not send camelCase variants.

Key response fields: TrainCapabilitiesVO contains schemaVersion, catalogVersion, catalogDigest, and entries. TrainJobDetailVO contains id, status, stage, progress, taskDisplayName, baseModel, trainingMethod, artifacts, actions, loss series, deployment references, timestamps, and error/status details.

const catalog = await client.training.capabilities();
const clusters = await client.training.clusterOptions();
const resource = await client.training.preview({
  clusterId: 1,
  resourceSpecId: "resource-spec-id",
});
const teachers = await client.training.knowledgeTeacherModels({
  studentModelId: "model-id",
  recipeId: "recipe-id",
});

const created = await client.training.create<{ jobId: string }>({
  clientToken: "training-request-001",
  displayName: "example-training",
  outputModelName: "example-output",
  recipeId: "recipe-id",
  recipeVersion: "recipe-version",
  baseModelRef: { type: "recipe_model", id: "model-id" },
  datasetRefs: [{ datasetId: "dataset-id", role: "train" }],
  placement: { clusterId: "1", resourceSpecId: "resource-spec-id" },
  params: {},
});
const page = await client.training.list({ pageNum: 1, pageSize: 20, status: "RUNNING" });
const detail = await client.training.get(created.jobId) as {
  artifacts?: Array<{ artifactId: string }>;
};
if (detail.artifacts?.length) {
  const download = await client.training.artifactDownloadUrl(
    detail.artifacts[0]!.artifactId,
  );
  const firstUrl = (download.urls ?? [download.url!])[0];
}
await client.training.cancel(created.jobId);

All methods use the shared authentication and typed error behavior described in Response Conventions and Retry and Security.