train_summarizer.py

from datasets import load_dataset
from transformers import T5ForConditionalGeneration, T5TokenizerFast, Trainer, TrainingArguments
import evaluate
import numpy as np

# Load dataset
dataset = load_dataset("cnn_dailymail", "3.0.0")

# Load tokenizer and model
tokenizer = T5TokenizerFast.from_pretrained("t5-small")
model = T5ForConditionalGeneration.from_pretrained("t5-small")

# Preprocess function
def preprocess_function(examples):
    inputs = ["summarize: " + doc for doc in examples["article"]]
    model_inputs = tokenizer(inputs, max_length=512, truncation=True)
    labels = tokenizer(text_target=examples["highlights"], max_length=128, truncation=True)
    model_inputs["labels"] = labels["input_ids"]
    return model_inputs

tokenized_datasets = dataset.map(preprocess_function, batched=True, remove_columns=["article", "highlights", "id"])

# Metrics
rouge = evaluate.load("rouge")

def compute_metrics(eval_pred):
    predictions, labels = eval_pred
    decoded_preds = tokenizer.batch_decode(predictions, skip_special_tokens=True)
    labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
    decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
    result = rouge.compute(predictions=decoded_preds, references=decoded_labels)
    return {k: v * 100 for k, v in result.items()}

# Training arguments
training_args = TrainingArguments(
    output_dir="./results",
    evaluation_strategy="epoch",
    learning_rate=3e-4,
    per_device_train_batch_size=2,
    per_device_eval_batch_size=2,
    num_train_epochs=1,
    save_strategy="epoch",
    predict_with_generate=True,
    push_to_hub=False
)

# Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets["train"].select(range(2000)),
    eval_dataset=tokenized_datasets["validation"].select(range(500)),
    tokenizer=tokenizer,
    compute_metrics=compute_metrics
)

trainer.train()
trainer.save_model("./t5-news-summarizer")
tokenizer.save_pretrained("./t5-news-summarizer")