from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, Seq2SeqTrainer, Seq2SeqTrainingArguments

model_checkpoint = "google/flan-t5-large"
output_dir = "./finetuned-flan-t5"

dataset = load_dataset("json", data_files={"train": "train_data.jsonl"})

tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)

def preprocess_function(examples):
    inputs = examples["input"]
    targets = examples["output"]
    model_inputs = tokenizer(inputs, max_length=512, truncation=True)
    labels = tokenizer(targets, max_length=128, truncation=True)
    model_inputs["labels"] = labels["input_ids"]
    return model_inputs

tokenized_datasets = dataset.map(preprocess_function, batched=True)

model = AutoModelForSeq2SeqLM.from_pretrained(model_checkpoint)

training_args = Seq2SeqTrainingArguments(
    output_dir=output_dir,
    evaluation_strategy="no",
    learning_rate=5e-5,
    per_device_train_batch_size=2,
    num_train_epochs=3,
    weight_decay=0.01,
    save_total_limit=2,
    push_to_hub=False
)

trainer = Seq2SeqTrainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets["train"]
)

trainer.train()

model.save_pretrained(output_dir)
tokenizer.save_pretrained(output_dir)