app.py

import gradio as gr
import torch
from transformers import AutoProcessor, Idefics3ForConditionalGeneration
import logging

logger = logging.getLogger(__name__)

class CustomModelChat:
    def __init__(self):
        self.model = None
        self.processor = None
        self.initialize_model()

    def initialize_model(self):
        try:
            model_id = "HuggingFaceTB/SmolVLM-Instruct"
            self.model = Idefics3ForConditionalGeneration.from_pretrained(
                model_id,
                device_map="auto",
                torch_dtype=torch.bfloat16
            )
            self.processor = AutoProcessor.from_pretrained(model_id)
            
            # Load your custom adapter
            adapter_path = "smolVLM_Essay_Knowledge_Distillation/smolvlm-instruct-trl-sft-ChartQA"
            self.model.load_adapter(adapter_path)
        except Exception as e:
            logger.error(f"Error initializing model: {e}")
            raise

    def process_chat_history(self, history, system_message):
        # Convert chat history to the format expected by the model
        messages = [{"role": "system", "content": system_message}]
        
        for user_msg, assistant_msg in history:
            if user_msg:
                messages.append({"role": "user", "content": user_msg})
            if assistant_msg:
                messages.append({"role": "assistant", "content": assistant_msg})
        
        return messages

    def generate_response(
        self,
        message,
        history,
        system_message,
        max_tokens=512,
        temperature=0.7,
        top_p=0.95,
        image=None
    ):
        try:
            messages = self.process_chat_history(history, system_message)
            messages.append({"role": "user", "content": message})

            # Prepare the chat template
            chat_input = self.processor.apply_chat_template(
                messages[1:],  # Exclude system message
                add_generation_prompt=True
            )

            # Handle image input if provided
            image_inputs = []
            if image is not None:
                if image.mode != 'RGB':
                    image = image.convert('RGB')
                image_inputs.append([image])

            # Prepare model inputs
            model_inputs = self.processor(
                text=chat_input,
                images=image_inputs if image_inputs else None,
                return_tensors="pt",
            ).to(self.model.device)

            # Generate response
            generated_ids = self.model.generate(
                **model_inputs,
                max_new_tokens=max_tokens,
                temperature=temperature,
                top_p=top_p,
                do_sample=True
            )

            # Trim and decode the response
            trimmed_ids = generated_ids[:, len(model_inputs.input_ids[0]):]
            response = self.processor.batch_decode(
                trimmed_ids,
                skip_special_tokens=True,
                clean_up_tokenization_spaces=False
            )[0]

            yield response

        except Exception as e:
            logger.error(f"Error generating response: {e}")
            yield f"Error: {str(e)}"

def create_chat_interface():
    chat_model = CustomModelChat()
    
    demo = gr.ChatInterface(
        chat_model.generate_response,
        additional_inputs=[
            gr.Textbox(value="You are a helpful assistant.", 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"),
            gr.Image(type="pil", label="Upload Image (optional)")
        ],
        title="Custom SmolVLM Chat",
        description="Chat interface using custom fine-tuned SmolVLM model"
    )
    
    return demo

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