Fine-tuning: customizing model behavior

This example demonstrates how to fine-tune a small language model (DistilGPT2) on a tiny instruction-response dataset so it adopts a simple clinic-assistant persona.

Requirements

This project requires Python 3.x and the libraries listed in requirements.txt.

Steps for running this example

1. Install dependencies:

python -m venv .venv
source .venv/bin/activate  # Windows cmd: .venv\Scripts\activate # Windows PowerShell: .venv\Scripts\Activate.ps1
pip install -r requirements.txt

2. Run the script:

python fine_tuning.py

Output

The script loads the dataset, tokenizes the examples, configures a Hugging Face Trainer, runs a short training loop, saves the updated model, and prints a sample generation.

[INFO] Loading tokenizer and model: distilgpt2
[INFO] Preparing dataset...
[INFO] Configuring trainer...
...
[INFO] Saving model to ./clinic-assistant-distilgpt2
[INFO] Testing fine-tuned model...
Generated Output:
Instruction: Help me book an appointment.
Response: Please arrive 15 minutes early to complete any necessary forms.