Create a Fine-Tuning Job
Start a fine-tuning job to create a custom model trained on your data.
require "openai"
client = OpenAI::Client.new
job = client.fine_tuning.jobs.create(
model: "gpt-4o-mini",
training_file: "file-abc123"
)
puts "Job ID: #{job.id}"
puts "Status: #{job.status}"
Before creating a fine-tuning job, you must first upload a training file using the Files API. Learn more in the file uploads guide. Fine-Tuning with Hyperparameters
Customize the training process with hyperparameters.
job = client.fine_tuning.jobs.create(
model: "gpt-4o-mini",
training_file: "file-abc123",
hyperparameters: {
n_epochs: 3,
batch_size: 8,
learning_rate_multiplier: 0.1
},
suffix: "my-custom-model"
)
puts "Job created: #{job.id}"
puts "Model name will be: #{job.fine_tuned_model || 'pending'}"
List Fine-Tuning Jobs
Retrieve all fine-tuning jobs for your organization.
page = client.fine_tuning.jobs.list(limit: 20)
# Access first job from page
job = page.data[0]
puts job.id
# Automatically fetch all jobs across pages
page.auto_paging_each do |job|
puts "Job: #{job.id} - Status: #{job.status}"
end
Retrieve a Fine-Tuning Job
Get detailed information about a specific fine-tuning job.
job = client.fine_tuning.jobs.retrieve("ft-AF1WoRqd3aJAHsqc9NY7iL8F")
puts "Job ID: #{job.id}"
puts "Model: #{job.model}"
puts "Status: #{job.status}"
puts "Created at: #{Time.at(job.created_at)}"
puts "Fine-tuned model: #{job.fine_tuned_model}"
puts "Trained tokens: #{job.trained_tokens}"
Monitor Job Events
Track the progress of a fine-tuning job by retrieving its events.
events = client.fine_tuning.jobs.list_events(
"ft-AF1WoRqd3aJAHsqc9NY7iL8F",
limit: 10
)
events.auto_paging_each do |event|
puts "[#{event.level}] #{Time.at(event.created_at)}: #{event.message}"
end
Upload training data
Use the Files API to upload your JSONL training file.file = client.files.create(
file: Pathname("training_data.jsonl"),
purpose: "fine-tune"
)
Create fine-tuning job
Start the fine-tuning process with your uploaded file.job = client.fine_tuning.jobs.create(
model: "gpt-4o-mini",
training_file: file.id
)
Monitor progress
Check job status and events periodically.job = client.fine_tuning.jobs.retrieve(job.id)
puts job.status
Use fine-tuned model
Once complete, use your custom model for completions.response = client.chat.completions.create(
model: job.fine_tuned_model,
messages: [{role: "user", content: "Hello!"}]
)
Cancel a Fine-Tuning Job
Stop a running fine-tuning job before completion.
job = client.fine_tuning.jobs.cancel("ft-AF1WoRqd3aJAHsqc9NY7iL8F")
puts "Job #{job.id} status: #{job.status}" # Should be "cancelled"
Pause and Resume Jobs
Temporarily pause and later resume a fine-tuning job.
job = client.fine_tuning.jobs.pause("ft-AF1WoRqd3aJAHsqc9NY7iL8F")
puts "Job paused: #{job.status}"
page = client.fine_tuning.jobs.list(limit: 20)
# Check if there are more pages
if page.next_page?
new_page = page.next_page
puts new_page.data[0].id
end
# Or iterate through all pages automatically
page.auto_paging_each do |job|
puts job.id
end
List Job Checkpoints
Retrieve checkpoints created during the fine-tuning process.
checkpoints = client.fine_tuning.jobs.checkpoints.list(
"ft-AF1WoRqd3aJAHsqc9NY7iL8F"
)
checkpoints.auto_paging_each do |checkpoint|
puts "Checkpoint: #{checkpoint.id}"
puts "Step: #{checkpoint.step_number}"
puts "Metrics: #{checkpoint.metrics}"
end
Fine-Tuning with Validation File
Provide a validation file to monitor overfitting during training.
# Upload validation file
validation_file = client.files.create(
file: Pathname("validation_data.jsonl"),
purpose: "fine-tune"
)
# Create job with validation
job = client.fine_tuning.jobs.create(
model: "gpt-4o-mini",
training_file: "file-abc123",
validation_file: validation_file.id,
hyperparameters: {
n_epochs: 3
}
)
puts "Training file: #{job.training_file}"
puts "Validation file: #{job.validation_file}"
Integration with Weights & Biases
Track fine-tuning metrics with Weights & Biases integration.
job = client.fine_tuning.jobs.create(
model: "gpt-4o-mini",
training_file: "file-abc123",
integrations: [
{
type: "wandb",
wandb: {
project: "my-fine-tuning-project",
name: "gpt-4o-mini-custom",
tags: ["production", "v1"]
}
}
]
)
Weights & Biases integration requires you to set up the integration in your OpenAI organization settings first.
Error Handling
Handle errors that may occur during fine-tuning.
begin
job = client.fine_tuning.jobs.create(
model: "gpt-4o-mini",
training_file: "file-abc123"
)
rescue OpenAI::Errors::APIConnectionError => e
puts "Server could not be reached: #{e.cause}"
rescue OpenAI::Errors::RateLimitError => e
puts "Rate limit exceeded, backing off"
rescue OpenAI::Errors::APIStatusError => e
puts "API error: #{e.status} - #{e.message}"
end
# Check job for errors after completion
job = client.fine_tuning.jobs.retrieve("ft-AF1WoRqd3aJAHsqc9NY7iL8F")
if job.error
puts "Job failed: #{job.error.message}"
puts "Error code: #{job.error.code}"
end