app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, AutoProcessor, AutoConfig, AutoModel
from PIL import Image
import logging
from transformers import BitsAndBytesConfig

# Setup logging
logging.basicConfig(level=logging.INFO)

class LLaVAPhiModel:
    def __init__(self, model_id="sagar007/Lava_phi"):
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        logging.info(f"Using device: {self.device}")
        
        # Initialize quantization config
        quantization_config = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_compute_dtype=torch.float16,
            bnb_4bit_use_double_quant=True,
            bnb_4bit_quant_type="nf4"
        )
        
        try:
            # Load model directly from Hugging Face Hub
            logging.info(f"Loading model from {model_id}...")
            self.model = AutoModelForCausalLM.from_pretrained(
                model_id,
                quantization_config=quantization_config,
                device_map="auto",
                torch_dtype=torch.bfloat16,
                trust_remote_code=True
            )
            self.tokenizer = AutoTokenizer.from_pretrained(model_id)
            
            # Set up padding token
            if self.tokenizer.pad_token is None:
                self.tokenizer.pad_token = self.tokenizer.eos_token
                self.model.config.pad_token_id = self.tokenizer.eos_token_id
                
            # Load CLIP model and processor
            logging.info("Loading CLIP model and processor...")
            self.processor = AutoProcessor.from_pretrained("openai/clip-vit-base-patch32")
            self.clip = AutoModel.from_pretrained("openai/clip-vit-base-patch32").to(self.device)
            
            # Store conversation history
            self.history = []
            
        except Exception as e:
            logging.error(f"Error initializing model: {str(e)}")
            raise
        
    def process_image(self, image):
        """Process image through CLIP"""
        with torch.no_grad():
            image_inputs = self.processor(images=image, return_tensors="pt")
            image_features = self.clip.get_image_features(
                pixel_values=image_inputs.pixel_values.to(self.device)
            )
            return image_features
        
    def generate_response(self, message, image=None):
        try:
            if image is not None:
                # Get image features
                image_features = self.process_image(image)
                
                # Format prompt
                prompt = f"human: <image>\n{message}\ngpt:"
                
                # Add context from history
                context = ""
                for turn in self.history[-3:]:
                    context += f"human: {turn[0]}\ngpt: {turn[1]}\n"
                
                full_prompt = context + prompt
                
                # Prepare text inputs
                inputs = self.tokenizer(
                    full_prompt, 
                    return_tensors="pt", 
                    padding=True,
                    truncation=True,
                    max_length=512
                )
                inputs = {k: v.to(self.device) for k, v in inputs.items()}
                
                # Add image features to inputs
                inputs["image_features"] = image_features
                
                # Generate response
                with torch.no_grad():
                    outputs = self.model.generate(
                        **inputs,
                        max_new_tokens=256,
                        min_length=20,
                        temperature=0.7,
                        do_sample=True,
                        top_p=0.9,
                        top_k=40,
                        repetition_penalty=1.5,
                        no_repeat_ngram_size=3,
                        use_cache=True,
                        pad_token_id=self.tokenizer.pad_token_id,
                        eos_token_id=self.tokenizer.eos_token_id
                    )
            else:
                # Text-only response
                prompt = f"human: {message}\ngpt:"
                context = ""
                for turn in self.history[-3:]:
                    context += f"human: {turn[0]}\ngpt: {turn[1]}\n"
                
                full_prompt = context + prompt
                inputs = self.tokenizer(
                    full_prompt,
                    return_tensors="pt",
                    padding=True,
                    truncation=True,
                    max_length=512
                )
                inputs = {k: v.to(self.device) for k, v in inputs.items()}
                
                with torch.no_grad():
                    outputs = self.model.generate(
                        **inputs,
                        max_new_tokens=150,
                        min_length=20,
                        temperature=0.6,
                        do_sample=True,
                        top_p=0.85,
                        top_k=30,
                        repetition_penalty=1.8,
                        no_repeat_ngram_size=4,
                        use_cache=True,
                        pad_token_id=self.tokenizer.pad_token_id,
                        eos_token_id=self.tokenizer.eos_token_id
                    )
            
            # Decode response
            response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
            
            # Clean up response
            if "gpt:" in response:
                response = response.split("gpt:")[-1].strip()
            if "human:" in response:
                response = response.split("human:")[0].strip()
            if "<image>" in response:
                response = response.replace("<image>", "").strip()
            
            # Update history
            self.history.append((message, response))
            
            return response
            
        except Exception as e:
            logging.error(f"Error generating response: {str(e)}")
            logging.error(f"Full traceback:", exc_info=True)
            return f"Error: {str(e)}"
    
    def clear_history(self):
        self.history = []
        return None

def create_demo():
    # Initialize model
    model = LLaVAPhiModel()
    
    with gr.Blocks(css="footer {visibility: hidden}") as demo:
        gr.Markdown(
            """
            # LLaVA-Phi Demo
            Chat with a vision-language model that can understand both text and images.
            """
        )
        
        chatbot = gr.Chatbot(height=400)
        with gr.Row():
            with gr.Column(scale=0.7):
                msg = gr.Textbox(
                    show_label=False,
                    placeholder="Enter text and/or upload an image",
                    container=False
                )
            with gr.Column(scale=0.15, min_width=0):
                clear = gr.Button("Clear")
            with gr.Column(scale=0.15, min_width=0):
                submit = gr.Button("Submit", variant="primary")
        
        image = gr.Image(type="pil", label="Upload Image (Optional)")
        
        def respond(message, chat_history, image):
            if not message and image is None:
                return chat_history
            
            response = model.generate_response(message, image)
            chat_history.append((message, response))
            return "", chat_history
        
        def clear_chat():
            model.clear_history()
            return None, None
        
        submit.click(
            respond,
            [msg, chatbot, image],
            [msg, chatbot],
        )
        
        clear.click(
            clear_chat,
            None,
            [chatbot, image],
        )
        
        msg.submit(
            respond,
            [msg, chatbot, image],
            [msg, chatbot],
        )
        
    return demo

if __name__ == "__main__":
    demo = create_demo()
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=True
    )