Fine-tuning a pre-trained large language model (LLM) involves adjusting the model parameters with a new dataset to customize its performance for specific tasks. Below is a step-by-step guide with a Python code snippet to help you fine-tune an LLM using the Hugging Face Transformers library, which is one of the most popular and well-supported frameworks for this purpose.

Step 1: Set Up the Environment

First, ensure you have the necessary libraries installed. You can install these using pip:

pip install transformers datasets torch

Step 2: Prepare the Dataset

For the purpose of this example, let's assume we are fine-tuning a model on a text classification task. We will use the datasets library to load a sample dataset. Here, we'll use the imdb movie review dataset, which is a popular binary sentiment classification dataset.

Here's how you can load and preprocess the data:

from datasets import load_dataset
# Load the dataset
dataset = load_dataset('imdb')
# Split the dataset into train and test sets
train_dataset = dataset['train']
test_dataset = dataset['test']

Step 3: Tokenize the Data

Next, we need to tokenize the text data using the tokenizer associated with the pre-trained model we plan to fine-tune. Here, we will use the bert-base-uncased model as an example:

from transformers import AutoTokenizer
# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')
# Tokenize the train and test sets
def tokenize_function(examples):
    return tokenizer(examples['text'], padding="max_length", truncation=True, max_length=512)
train_dataset =, batched=True)
test_dataset =, batched=True)
# Set format for PyTorch
train_dataset.set_format(type='torch', columns=['input_ids', 'attention_mask', 'label'])
test_dataset.set_format(type='torch', columns=['input_ids', 'attention_mask', 'label'])

Step 4: Load the Pre-trained Model

Now that our data is tokenized, let's load the pre-trained model. We'll use the BertForSequenceClassification model from transformers:

from transformers import AutoModelForSequenceClassification
# Load the model
model = AutoModelForSequenceClassification.from_pretrained('bert-base-uncased', num_labels=2)

Step 5: Define the Training Arguments

We'll now define the training arguments using the TrainingArguments class. These arguments control the behavior of the training loop.

from transformers import TrainingArguments
training_args = TrainingArguments(
    num_train_epochs=3,             # Number of training epochs
    per_device_train_batch_size=16, # Batch size for training
    per_device_eval_batch_size=16# Batch size for evaluation
    warmup_steps=500,               # Number of warmup steps for learning rate scheduler
    weight_decay=.01,              # Strength of weight decay
    logging_dir='./logs',           # Directory for storing logs

Step 6: Create the Trainer

Using the Trainer API makes it convenient to handle the training process. We provide the model, arguments, datasets, and evaluation metrics to the Trainer.

from transformers import Trainer
def compute_metrics(p):
    from sklearn.metrics import accuracy_score, precision_recall_fscore_support
    preds = p.predictions.argmax(-1)
    labels = p.label_ids
    precision, recall, f1, _ = precision_recall_fscore_support(labels, preds, average='binary')
    acc = accuracy_score(labels, preds)
    return {
        'accuracy': acc,
        'f1': f1,
        'precision': precision,
        'recall': recall
trainer = Trainer(
    model=model,                         # The model to be trained
    args=training_args,                  # Training arguments
    train_dataset=train_dataset,         # Training dataset
    eval_dataset=test_dataset,           # Evaluation dataset
    compute_metrics=compute_metrics      # Function to compute metrics

Step 7: Train the Model

Finally, you can start the training process:


Step 8: Evaluate the Model

After training, you can evaluate the model on the test set:

evaluation_results = trainer.evaluate()

Step 9: Save the Model

Once the model is fine-tuned, save it for later use:



This step-by-step guide provided you with an overview of fine-tuning a pre-trained large language model using the transformers library. You learned how to set up the environment, prepare and tokenize your dataset, define training arguments, and train and evaluate the model using the Trainer API.