training/training.py

from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer
from datasets import load_dataset, DatasetDict
import torch

# Load DistilBERT tokenizer and model
model_name = "distilbert-base-uncased"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=3)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
print("Current Device:", torch.cuda.current_device())
# Load dataset
dataset = load_dataset("Faith1712/allsides_text_proper_truncated")

# Split dataset into train and eval sets (90% train, 10% eval)
dataset = dataset["train"].train_test_split(test_size=0.1)

# Tokenization function
def tokenize_function(example):
    return tokenizer(example["text"], padding="max_length", truncation=True, max_length=512)

# Apply tokenization
tokenized_datasets = dataset.map(tokenize_function, batched=True)

# Define training arguments
training_args = TrainingArguments(
    output_dir="./distilbert-bias-detector",
    evaluation_strategy="epoch",  # Evaluates at end of each epoch
    save_strategy="epoch",
    per_device_train_batch_size=16,
    per_device_eval_batch_size=16,
    num_train_epochs=3,
    weight_decay=0.01,
    logging_dir="./logs",
    logging_steps=500,
)

# Define Trainer with both train and evaluation datasets
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets["train"],
    eval_dataset=tokenized_datasets["test"],  # Pass the test split for evaluation
    tokenizer=tokenizer,
)

# Start Training
trainer.train()