Implementation:Openai Openai node FineTuning Checkpoints

Knowledge Sources	Openai_Openai_node
Domains	SDK, Fine-Tuning
Last Updated	2026-02-15 12:00 GMT

Overview

The Checkpoints class provides a paginated listing endpoint for fine-tuning job checkpoints, which represent intermediate model snapshots produced during a fine-tuning run.

Description

The Checkpoints class extends APIResource and exposes a single list method that retrieves all checkpoints associated with a given fine-tuning job. Each checkpoint captures a model snapshot at a particular training step, along with metrics such as training loss, validation loss, and mean token accuracy. Checkpoints are usable as models in other API endpoints.

The method returns a PagePromise wrapping a CursorPage of FineTuningJobCheckpoint objects, enabling automatic cursor-based pagination via for await loops. The module also exports the FineTuningJobCheckpoint interface describing the checkpoint object shape, including a nested Metrics interface that captures training and validation statistics at each step.

The pagination parameters follow the standard CursorPageParams pattern used throughout the SDK, accepting optional after and limit fields.

Usage

Use this resource when you need to inspect the intermediate training progress of a fine-tuning job, compare metrics across training steps, or select a specific checkpoint model for inference rather than the final model.

Code Reference

Source Location

• Repository: openai-node
• File: src/resources/fine-tuning/jobs/checkpoints.ts

Signature

export class Checkpoints extends APIResource {
  list(
    fineTuningJobID: string,
    query?: CheckpointListParams | null | undefined,
    options?: RequestOptions,
  ): PagePromise<FineTuningJobCheckpointsPage, FineTuningJobCheckpoint>;
}

export interface FineTuningJobCheckpoint {
  id: string;
  created_at: number;
  fine_tuned_model_checkpoint: string;
  fine_tuning_job_id: string;
  metrics: FineTuningJobCheckpoint.Metrics;
  object: 'fine_tuning.job.checkpoint';
  step_number: number;
}

export namespace FineTuningJobCheckpoint {
  export interface Metrics {
    full_valid_loss?: number;
    full_valid_mean_token_accuracy?: number;
    step?: number;
    train_loss?: number;
    train_mean_token_accuracy?: number;
    valid_loss?: number;
    valid_mean_token_accuracy?: number;
  }
}

Import

import OpenAI from 'openai';

I/O Contract

Inputs

Name	Type	Required	Description
fineTuningJobID	string	Yes	The identifier of the fine-tuning job whose checkpoints to list.
query.after	string	No	Cursor for pagination; return results after this checkpoint ID.
query.limit	number	No	Maximum number of checkpoints to return per page.
options	RequestOptions	No	Additional request configuration (headers, timeout, etc.).

Outputs

Name	Type	Description
id	string	Unique checkpoint identifier, usable as a model reference.
created_at	number	Unix timestamp (seconds) when the checkpoint was created.
fine_tuned_model_checkpoint	string	Name of the fine-tuned checkpoint model.
fine_tuning_job_id	string	ID of the parent fine-tuning job.
metrics	Metrics	Training and validation metrics at this step (loss, accuracy).
object	'fine_tuning.job.checkpoint'	Object type discriminator.
step_number	number	The training step at which the checkpoint was saved.

Usage Examples

import OpenAI from 'openai';

const client = new OpenAI();

// Automatically fetches more pages as needed.
for await (const checkpoint of client.fineTuning.jobs.checkpoints.list(
  'ft-AF1WoRqd3aJAHsqc9NY7iL8F',
)) {
  console.log(`Step ${checkpoint.step_number}: loss=${checkpoint.metrics.train_loss}`);
}

Related Pages

• Environment:Openai_Openai_node_Node_20_Runtime