Fine-Tuning GPT-4o for Domain-Specific Tasks on Azure

While GPT-4o excels at general tasks, fine-tuning unlocks exceptional performance for domain-specific applications. Azure OpenAI’s fine-tuning service now supports GPT-4o, enabling custom models that maintain safety guardrails while adapting to your specific use cases.

Preparing Training Data

Quality training data is critical. Structure your examples to demonstrate the exact behavior you want:

import json
from pathlib import Path

def prepare_training_data(examples: list[dict]) -> str:
    """Format examples for Azure OpenAI fine-tuning."""
    training_lines = []

    for ex in examples:
        training_example = {
            "messages": [
                {"role": "system", "content": ex["system_prompt"]},
                {"role": "user", "content": ex["user_input"]},
                {"role": "assistant", "content": ex["expected_output"]}
            ]
        }
        training_lines.append(json.dumps(training_example))

    return "\n".join(training_lines)

# Example: Legal document analysis
legal_examples = [
    {
        "system_prompt": "You are a legal document analyst. Extract key terms and obligations.",
        "user_input": "Analyze this contract clause: 'The Licensee shall pay...'",
        "expected_output": '{"obligation_type": "payment", "party": "Licensee", ...}'
    },
    # 50-100 high-quality examples recommended
]

training_data = prepare_training_data(legal_examples)
Path("training_data.jsonl").write_text(training_data)

Initiating Fine-Tuning

Use the Azure OpenAI SDK to manage the fine-tuning process:

from openai import AzureOpenAI

client = AzureOpenAI(
    api_key=api_key,
    api_version="2024-12-01-preview",
    azure_endpoint=endpoint
)

# Upload training file
with open("training_data.jsonl", "rb") as f:
    training_file = client.files.create(file=f, purpose="fine-tune")

# Create fine-tuning job
job = client.fine_tuning.jobs.create(
    training_file=training_file.id,
    model="gpt-4o-2024-08-06",
    hyperparameters={
        "n_epochs": 3,
        "batch_size": 4,
        "learning_rate_multiplier": 0.5
    },
    suffix="legal-analyst"
)

print(f"Fine-tuning job created: {job.id}")

Evaluation and Deployment

After training completes, evaluate on a held-out test set before deploying. Monitor the fine-tuned model’s performance against the base model to ensure improvements justify the additional cost.