scripts/train.py

from transformers import RobertaTokenizer, RobertaForSequenceClassification, Trainer, TrainingArguments
from datasets import load_dataset

# 1. Preparar o Dataset
def load_data_from_csv(csv_file):
    dataset = load_dataset("csv", data_files=csv_file)
    return dataset['train']

# 2. Configurar o Tokenizer e Modelo
def get_model_and_tokenizer():
    model_name = "microsoft/codebert-base"
    tokenizer = RobertaTokenizer.from_pretrained(model_name)
    model = RobertaForSequenceClassification.from_pretrained(model_name, num_labels=2)  # 2 classes: safe/unsafe
    return model, tokenizer

# 3. Tokenizar os Dados
def tokenize_function(example, tokenizer):
    return tokenizer(example['content'], truncation=True, padding="max_length", max_length=512)

# 4. Treinar o Modelo
def train_model(dataset, tokenizer, model):
    tokenized_data = dataset.map(lambda x: tokenize_function(x, tokenizer), batched=True)
    training_args = TrainingArguments(
        output_dir="./results",
        evaluation_strategy="epoch",
        save_strategy="epoch",
        learning_rate=2e-5,
        num_train_epochs=3,
        per_device_train_batch_size=16,
        per_device_eval_batch_size=16,
        warmup_steps=500,
        weight_decay=0.01,
        logging_dir="./logs",
        logging_steps=10,
    )
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=tokenized_data,
        eval_dataset=tokenized_data,
        tokenizer=tokenizer,
    )
    trainer.train()

if __name__ == "__main__":
    # Carregar Dados e Modelo
    dataset = load_data_from_csv("code_analysis_dataset.csv")
    model, tokenizer = get_model_and_tokenizer()

    # Treinar Modelo
    train_model(dataset, tokenizer, model)
    print("[SUCCESS] Model trained!")