Practical Fine-tuning

Data Preparation

from datasets import Dataset

def prepare_training_data(instructions, model_name):
    """Format data for instruction tuning."""
    formatted_data = []

    for inst in instructions:
        text = f"""### Instruction:
{inst['instruction']}

### Response:
{inst['response']}"""

        formatted_data.append({"text": text})

    dataset = Dataset.from_list(formatted_data)

    def tokenize(examples):
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        return tokenizer(examples["text"], truncation=True, max_length=512)

    return dataset.map(tokenize, batched=True)

Training Configuration

from transformers import TrainingArguments

training_args = TrainingArguments(
    output_dir="./output",
    num_train_epochs=3,
    per_device_train_batch_size=4,
    gradient_accumulation_steps=4,
    learning_rate=2e-4,
    weight_decay=0.01,
    warmup_ratio=0.03,
    lr_scheduler_type="cosine",
    fp16=True,
    logging_steps=10,
    save_steps=500,
    evaluation_strategy="steps",
    eval_steps=500,
    load_best_model_at_end=True,
)

Best Practices

Summary

Successful fine-tuning requires careful data preparation, appropriate hyperparameters, and systematic evaluation. Start simple and iterate.

Next: We'll explore AI ethics and bias.