How to Intermediate · 4 min read

How to fine-tune GPT-3.5 with OpenAI API

Quick answer

You cannot fine-tune GPT-3.5 directly via the OpenAI API as of 2026-04; instead, use OpenAI's fine-tuning endpoints on supported base models like gpt-3.5-turbo-0613 or gpt-4o. Prepare your training data in JSONL format, upload it, create a fine-tune job, and then use the fine-tuned model for inference via the API.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install openai>=1.0

Setup

Install the official OpenAI Python SDK and set your API key as an environment variable for secure access.

bash

pip install openai>=1.0

Step by step

Prepare your training data in JSONL format with prompt and completion fields, upload it, create a fine-tune job, and then query the fine-tuned model.

python

import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# Step 1: Upload training data
# training_data.jsonl example line: {"prompt": "Translate English to French: 'Hello'", "completion": " Bonjour"}

upload_response = client.files.create(
    file=open("training_data.jsonl", "rb"),
    purpose="fine-tune"
)
file_id = upload_response.id
print(f"Uploaded file ID: {file_id}")

# Step 2: Create fine-tune job
fine_tune_response = client.fine_tunes.create(
    training_file=file_id,
    model="gpt-3.5-turbo-0613"
)
fine_tune_id = fine_tune_response.id
print(f"Fine-tune job created with ID: {fine_tune_id}")

# Step 3: Poll fine-tune job status (simplified)
import time

while True:
    status_response = client.fine_tunes.get(id=fine_tune_id)
    status = status_response.status
    print(f"Fine-tune status: {status}")
    if status in ["succeeded", "failed"]:
        break
    time.sleep(10)

# Step 4: Use fine-tuned model
if status == "succeeded":
    fine_tuned_model = status_response.fine_tuned_model
    completion = client.chat.completions.create(
        model=fine_tuned_model,
        messages=[{"role": "user", "content": "Translate English to French: 'Good morning'"}]
    )
    print("Response:", completion.choices[0].message.content)
else:
    print("Fine-tuning failed.")

output

Uploaded file ID: file-abc123xyz
Fine-tune job created with ID: ft-xyz789abc
Fine-tune status: running
Fine-tune status: succeeded
Response: Bonjour

Common variations

You can fine-tune other supported base models like gpt-4o. Use async calls or streaming for large jobs. The OpenAI SDK supports these variations with similar method calls.

python

import asyncio
import os
from openai import OpenAI

async def fine_tune_async():
    client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
    upload = await client.files.acreate(file=open("training_data.jsonl", "rb"), purpose="fine-tune")
    ft = await client.fine_tunes.acreate(training_file=upload.id, model="gpt-4o")
    print(f"Async fine-tune job ID: {ft.id}")

asyncio.run(fine_tune_async())

output

Async fine-tune job ID: ft-async123

Troubleshooting

If you get a 400 Bad Request, check your JSONL formatting and ensure prompt and completion fields are correct.
If fine-tuning fails, verify your training data size and quality.
Use client.fine_tunes.list_events(id=fine_tune_id) to debug errors.

✅

Key Takeaways

Fine-tuning GPT-3.5 requires using supported base models like gpt-3.5-turbo-0613 via OpenAI's fine-tune API.
Prepare training data in JSONL format with prompt-completion pairs and upload it before creating a fine-tune job.
Poll the fine-tune job status to know when the model is ready, then use the fine-tuned model for inference.
Use the official OpenAI Python SDK v1+ with environment variable API keys for secure and up-to-date integration.
Check fine-tune job events and logs for troubleshooting common errors like formatting or data issues.

Verified 2026-04 · gpt-3.5-turbo-0613, gpt-4o

Verify ↗