Principle:Openai Openai node Fine Tuning Job Creation
| Knowledge Sources | |
|---|---|
| Domains | Fine_Tuning, Model_Training |
| Last Updated | 2026-02-15 00:00 GMT |
Overview
A principle for creating a fine-tuning job that trains a base model on custom data using specified training methods and hyperparameters.
Description
Fine-Tuning Job Creation initiates the actual model training process. It references a processed training file, selects a base model, and optionally configures hyperparameters and training methods. OpenAI supports multiple training methods: supervised (standard SFT), dpo (Direct Preference Optimization), and reinforcement (RL-based fine-tuning).
The job enters an asynchronous processing pipeline: validating_files → queued → running → succeeded (or failed / cancelled). Upon success, the job produces a fine-tuned model ID.
Usage
Use this principle after uploading and verifying training data. This step initiates training; monitor progress via the Job Progress Monitoring principle.
Theoretical Basis
Job creation follows an Asynchronous Job Submission pattern:
function createFineTuningJob(trainingFileId, model, options):
job = await api.post('/fine_tuning/jobs', {
training_file: trainingFileId,
model: model, // e.g., 'gpt-4o-mini'
suffix: options.suffix, // Custom model name suffix
method: options.method, // supervised | dpo | reinforcement
hyperparameters: options.hyperparameters,
})
// Job starts asynchronous processing
// job.status: 'validating_files'
// job.fine_tuned_model: null (until succeeded)
return job