Lesson 9: Fine-tune on OpenAI — GPT-4o-mini & GPT-4o

Fine-tuning LLM: The Art of AI Tuning

Part 4: Fine-tuning on OpenAI & other Platforms

xdev.asia

Introduction

OpenAI is the most popular fine-tuning platform — large ecosystem, good documentation, but inference is more expensive than Google. This article compares hands-on.

---

1. OpenAI Fine-tuning Workflow

from openai import OpenAI
client = OpenAI()

# Upload dataset
file = client.files.create(file=open("data.jsonl","rb"), purpose="fine-tune")

# Create fine-tuning job
job = client.fine_tuning.jobs.create(
    training_file=file.id,
    model="gpt-4o-mini-2024-07-18",
    hyperparameters={"n_epochs": 3}
)

# Monitor
while True:
    status = client.fine_tuning.jobs.retrieve(job.id)
    print(f"Status: {status.status}")
    if status.status in ["succeeded", "failed"]:
        break
    time.sleep(60)

# Use fine-tuned model
response = client.chat.completions.create(
    model=status.fine_tuned_model,
    messages=[{"role": "user", "content": "Test query"}]
)

2. OpenAI vs Google — When to use what?

	OpenAI	Google Vertex AI
Training cost	$3.00/1M tokens (mini)	~$0.40/1M tokens
Inference cost	2x base price	Equals base price
Ecosystem	Biggest	In development
Ease of use	Very easy	Need GCP setup
Best for	Prototype, small scale	Production, large scale

---

Summary

• OpenAI fine-tuning is simple: upload → create job → wait → use
• Inference is 2x more expensive than the base model — need to calculate carefully for production
• Use OpenAI for prototype, turn to Google for production scale

Exercises

1. Fine-tune GPT-4o-mini with the same dataset used for Gemini
2. Compare output quality: Gemini FT vs OpenAI FT
3. Cost comparison: training + 30 days inference
4. Conclusion: which provider is suitable for your use case?