t5_urdu_translation

import torch
from datasets import load_dataset
from transformers import T5Tokenizer, T5ForConditionalGeneration, Trainer, TrainingArguments

# Load dataset (replace 'your_dataset' with your actual dataset path or Hugging Face dataset name)
dataset = load_dataset('csv', data_files={'train': 'train.csv', 'validation': 'validation.csv'})

# Preprocess dataset
def preprocess_function(examples):
    inputs = ["translate English to Urdu: " + ex for ex in examples["English"]]
    targets = examples["Urdu"]
    model_inputs = tokenizer(inputs, max_length=512, truncation=True)
    labels = tokenizer(targets, max_length=512, truncation=True).input_ids
    model_inputs["labels"] = labels
    return model_inputs

# Load T5 tokenizer and model
model_name = "t5-small"
tokenizer = T5Tokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name)

# Tokenize datasets
tokenized_datasets = dataset.map(preprocess_function, batched=True)

# Define training arguments
training_args = TrainingArguments(
    output_dir="./t5_urdu_translation",
    evaluation_strategy="epoch",
    learning_rate=5e-5,
    per_device_train_batch_size=16,
    per_device_eval_batch_size=16,
    num_train_epochs=3,
    weight_decay=0.01,
    save_total_limit=2,
    predict_with_generate=True,
    logging_dir="./logs",
)

# Define Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets["train"],
    eval_dataset=tokenized_datasets["validation"],
    tokenizer=tokenizer,
)

# Train model
trainer.train()

# Save model
trainer.save_model("./t5_urdu_translation")
tokenizer.save_pretrained("./t5_urdu_translation")