Duckbot.py

import pandas as pd
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, Trainer, TrainingArguments
from Features.chat_interface import start_chat_interface 
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, Trainer, TrainingArguments
from Features.chat_interface import start_chat_interface
import start_chat_interface, generate_response  # Import needed functions
import os

# --- Step 1: Prepare Your Data ---
data = pd.read_csv("your_chatbot_data.csv")  # Replace with your dataset file
data = data[["user_input", "chatbot_response"]]  # Adjust column names if needed
data_path = "data/chatbot_data.csv"
data = pd.read_csv(data_path) 

# --- Step 2: Choose a Pre-trained Model ---
model_name = "microsoft/DialoGPT-medium"  # A good starting point, experiment with others! 
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

# --- Step 3: Tokenize the Data ---
def preprocess(examples):
    inputs = [ex + tokenizer.eos_token for ex in examples["user_input"]]
    targets = examples["chatbot_response"]
    model_inputs = tokenizer(inputs, max_length=128, truncation=True, padding=True)

    with tokenizer.as_target_tokenizer():
        model_inputs["labels"] = tokenizer(targets, max_length=128, truncation=True, padding=True).input_ids

    return model_inputs

tokenized_data = data.apply(preprocess, axis=1) 

# --- Step 4: Fine-Tune the Model ---
training_args = TrainingArguments(
    "my-chatbot",  # Output folder name 
    evaluation_strategy="steps",
    learning_rate=2e-5,
    per_device_train_batch_size=8,
    per_device_eval_batch_size=8,
    num_train_epochs=3, 
    weight_decay=0.01,
    save_steps=500,  # Save model checkpoint every 500 steps
    push_to_hub=True,  # Push the model to your Hugging Face Hub
)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_data,
    eval_dataset=tokenized_data,  
)

trainer.train()

# --- Step 5: Use Your Fine-Tuned Model (After Training) ---
def generate_response(user_input):
  input_ids = tokenizer.encode(user_input + tokenizer.eos_token, return_tensors='pt')
  output_sequences = model.generate(input_ids=input_ids)
  response = tokenizer.decode(output_sequences[0], skip_special_tokens=True) 
  return response