Skip to main content

Start Training

Start a new model fine-tuning job using the Knowledgator TypeScript SDK.


Configuration Template​

When starting a training job, specify your target base model, training dataset, and optional hyperparameters:

  • model_id ("<BASE_MODEL_ID>"): The base model to fine-tune. Options include 'comprehend-it' (zero-shot text classification and NLI), 'gliclass-x-base' (general entity/token classification), or any model identifier returned by kg.finetuning.listModels().
  • dataset_id ("<DATASET_ID>"): The unique UUID of your training dataset (e.g. '6babd48a-2256-4466-b16b-9fd45ba59688'), obtained from kg.datasets.list() or after uploading via kg.datasets.create().
  • hyperparameters: Optional dictionary of training arguments (learning_rate, batch_size, epochs, seed).
import { createClient } from '@knowledgator/typescript-sdk';

const kg = createClient();

const training = await kg.finetuning.startTraining({
model_id: '<BASE_MODEL_ID>',
dataset_id: '<DATASET_ID>',
hyperparameters: {
learning_rate: 0.0001,
batch_size: 16,
epochs: 5
}
});

Concrete Example​

Fine-tuning the comprehend-it base model using an uploaded dataset:

import { createClient } from '@knowledgator/typescript-sdk';

const kg = createClient();

const training = await kg.finetuning.startTraining({
model_id: 'comprehend-it',
dataset_id: '6babd48a-2256-4466-b16b-9fd45ba59688',
hyperparameters: {
learning_rate: 0.0001,
batch_size: 16,
epochs: 5
}
});
console.log(training.id, training.status);

Expected Output​

78d379da228a4985ba19100f53245b79 pending

Result Object Structure​

The SDK provides normalized properties with automatic fallbacks (training_id vs id, parent_model_id vs model_id, dataset_id vs nested raw.training.dataset_id):

{
"id": "78d379da228a4985ba19100f53245b79",
"model_id": "comprehend-it",
"dataset_id": "6babd48a-2256-4466-b16b-9fd45ba59688",
"status": "pending",
"type": "model",
"created_at": "2024-02-15T09:00:00Z"
}