app.py

'''
import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import AutoPeftModelForCausalLM
import torch

# Load the model and tokenizer
def load_model():
    # base_model_name = "unsloth/llama-3.2-1b-instruct-bnb-4bit"  # Replace with your base model name
    lora_model_name = "sreyanghosh/lora_model"  # Replace with your LoRA model path
    # tokenizer = AutoTokenizer.from_pretrained(base_model_name)
    # model = AutoModelForCausalLM.from_pretrained(
    #     base_model_name, 
    #     device_map="auto" if torch.cuda.is_available() else None,
    #     load_in_8bit=not torch.cuda.is_available(),
    # )
    # model = PeftModel.from_pretrained(model, lora_model_name)
    
    model = AutoPeftModelForCausalLM.from_pretrained(
        lora_model_name, # YOUR MODEL YOU USED FOR TRAINING
        load_in_4bit = False, # False
    )
    tokenizer = AutoTokenizer.from_pretrained(lora_model_name)
    model.eval()
    return tokenizer, model

tokenizer, model = load_model()

# Define the respond function
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # Prepare the conversation history
    messages = [{"role": "system", "content": system_message}]
    for user_input, bot_response in history:
        if user_input:
            messages.append({"role": "user", "content": user_input})
        if bot_response:
            messages.append({"role": "assistant", "content": bot_response})
    messages.append({"role": "user", "content": message})

    # Format the input for the model
    conversation_text = "\n".join(
        f"{msg['role']}: {msg['content']}" for msg in messages
    )
    inputs = tokenizer(conversation_text, return_tensors="pt", truncation=True)
    
    # Generate the model's response
    outputs = model.generate(
        inputs.input_ids,
        max_length=len(inputs.input_ids[0]) + max_tokens,
        temperature=temperature,
        top_p=top_p,
        pad_token_id=tokenizer.eos_token_id,
    )
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Extract the new response
    new_response = response[len(conversation_text):].strip()
    yield new_response

# Gradio app configuration
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)

if __name__ == "__main__":
    demo.launch()
'''

import gradio as gr
from transformers import AutoTokenizer
from peft import AutoPeftModelForCausalLM
import torch

# Load the model and tokenizer
def load_model():
    lora_model_name = "sreyanghosh/lora_model"  # Replace with your LoRA model path

    # Try loading without 4-bit quantization
    model = AutoPeftModelForCausalLM.from_pretrained(
        lora_model_name,
        torch_dtype=torch.float32,  # Ensure no low-bit quantization
        device_map="auto" if torch.cuda.is_available() else None,  # Use standard device mapping
        load_in_4bit=False,  # Redundant, but safe to explicitly specify
        )

    tokenizer = AutoTokenizer.from_pretrained(lora_model_name)

    if tokenizer.pad_token_id is None:
        tokenizer.pad_token_id = tokenizer.eos_token_id

    model.eval()
    device = "cuda" if torch.cuda.is_available() else "cpu"
    model = model.to(device)

    return tokenizer, model


# Define the respond function
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # Prepare the conversation history
    messages = [{"role": "system", "content": system_message}]
    for user_input, bot_response in history:
        if user_input:
            messages.append({"role": "user", "content": user_input})
        if bot_response:
            messages.append({"role": "assistant", "content": bot_response})
    messages.append({"role": "user", "content": message})

    # Format the input for the model
    conversation_text = "\n".join(
        f"{msg['role']}: {msg['content']}" for msg in messages
    )
    inputs = tokenizer(conversation_text, return_tensors="pt", truncation=True).to(model.device)
    
    # Generate the model's response
    outputs = model.generate(
        inputs.input_ids,
        max_length=len(inputs.input_ids[0]) + max_tokens,
        temperature=temperature,
        top_p=top_p,
        pad_token_id=tokenizer.eos_token_id,
        eos_token_id=tokenizer.eos_token_id,
    )
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Extract the new response
    new_response = response.split("assistant:")[-1].strip()
    yield new_response

# Gradio app configuration
demo = gr.ChatInterface(
    fn=respond,
    chatbot=gr.Chatbot(label="Assistant"),  # Use a Gradio Chatbot component
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)


if __name__ == "__main__":
    demo.launch()