app.py

import os
import pandas as pd
from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, Trainer, TrainingArguments
from transformers import pipeline
from sklearn.model_selection import train_test_split

def load_and_preprocess_data():
    dataset = load_dataset('tahiryaqoob/BISELahore')
    train_dataset, val_dataset = train_test_split(dataset['train'], test_size=0.2, random_state=42)
    
    print(f"Training samples: {len(train_dataset)}")
    print(f"Validation samples: {len(val_dataset)}")
    return train_dataset, val_dataset

#Preprocess the data to format for fine-tunin
def preprocess_function(examples, tokenizer):
    inputs = tokenizer(examples['question'], padding="max_length", truncation=True, max_length=128)
    targets = tokenizer(examples['answer'], padding="max_length", truncation=True, max_length=128)
    inputs['labels'] = targets['input_ids']
    return inputs
    
def fine_tune_model(train_dataset, val_dataset):
    model_name = "distilbert-base-uncased"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

    train_dataset = train_dataset.map(lambda x: preprocess_function(x, tokenizer), batched=True)
    val_dataset = val_dataset.map(lambda x: preprocess_function(x, tokenizer), batched=True)

    training_args = TrainingArguments(
        output_dir="./distilbert_finetuned",
        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=200,
        evaluation_strategy="epoch",
        save_strategy="epoch",
    )

    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset,
        eval_dataset=val_dataset,
        tokenizer=tokenizer,
    )

    trainer.train()

    model.save_pretrained("./distilbert_finetuned")
    tokenizer.save_pretrained("./distilbert_finetuned")
    print("Model fine-tuned and saved successfully.")

#Create a chatbot inference pipeline using the fine-tuned model
def chatbot_inference():
    model_name = "./distilbert_finetuned"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

    chatbot = pipeline("text2text-generation", model=model, tokenizer=tokenizer)
    return chatbot

#Run inference to test chatbot functionality
def run_inference():
    chatbot = chatbot_inference()
    user_input = input("Ask a question: ")
    response = chatbot(user_input)
    print("Bot Response:", response[0]['generated_text'])

#Main function to train or serve the chatbot
def main():
    train_dataset, val_dataset = load_and_preprocess_data()

    if not os.path.exists("./distilbert_finetuned"):
        fine_tune_model(train_dataset, val_dataset)
    else:
        print("Fine-tuned model already exists. Skipping fine-tuning.")

    run_inference()

if __name__ == "__main__":
    main()