app.py

import sys
sys.path.append('./LLAUS')

from transformers import AutoTokenizer, AutoModelForCausalLM, AutoConfig
import torch
from llava import LlavaLlamaForCausalLM
from llava.conversation import conv_templates
from llava.utils import disable_torch_init
from transformers import CLIPVisionModel, CLIPImageProcessor, StoppingCriteria
from PIL import Image
from torch.cuda.amp import autocast
import gradio as gr
import spaces
from peft import prepare_model_for_int8_training, LoraConfig, get_peft_model
import os
from transformers import AutoProcessor, AutoModel
import torch.nn.functional as F

#---------------------------------
#++++++++     Model     ++++++++++
#---------------------------------

DEFAULT_IMAGE_TOKEN = "<image>"
DEFAULT_IMAGE_PATCH_TOKEN = "<im_patch>"
DEFAULT_IM_START_TOKEN = "<im_start>"
DEFAULT_IM_END_TOKEN = "<im_end>"

def patch_config(config_path):
    """Applies necessary patches to the model config."""
    patch_dict = {
        "use_mm_proj": True,
        "mm_vision_tower": "openai/clip-vit-large-patch14",
        "mm_hidden_size": 1024
    }
    cfg = AutoConfig.from_pretrained(config_path)
    if not hasattr(cfg, "mm_vision_tower"):
        print(f'`mm_vision_tower` not found in `{config_path}`, applying patch and save to disk.')
        for k, v in patch_dict.items():
            setattr(cfg, k, v)
        cfg.save_pretrained(config_path)

def load_llava_model():
    """Loads and initializes the LLaVA model."""
    model_name = "Baron-GG/LLaVA-Med" # Change this to your model if you uploaded a new one
    disable_torch_init()
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    patch_config(model_name)
    model = LlavaLlamaForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16).cuda()
    model.model.requires_grad_(False)

    image_processor = CLIPImageProcessor.from_pretrained(model.config.mm_vision_tower, torch_dtype=torch.float16)

    model.config.use_cache = False
    model.config.tune_mm_mlp_adapter = False
    model.config.freeze_mm_mlp_adapter = False
    model.config.mm_use_im_start_end = True

    mm_use_im_start_end = getattr(model.config, "mm_use_im_start_end", False)
    tokenizer.add_tokens([DEFAULT_IMAGE_PATCH_TOKEN], special_tokens=True)
    if mm_use_im_start_end:
        tokenizer.add_tokens([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN], special_tokens=True)

    vision_tower = model.model.vision_tower[0]
    vision_tower.to(device='cuda', dtype=torch.float16)
    vision_config = vision_tower.config
    vision_config.im_patch_token = tokenizer.convert_tokens_to_ids([DEFAULT_IMAGE_PATCH_TOKEN])[0]
    vision_config.use_im_start_end = mm_use_im_start_end
    if mm_use_im_start_end:
        vision_config.im_start_token, vision_config.im_end_token = tokenizer.convert_tokens_to_ids([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN])
    image_token_len = (vision_config.image_size // vision_config.patch_size) ** 2
    
    model = prepare_model_for_int8_training(model)
    lora_config = LoraConfig(
        r=64,
        lora_alpha=16,
        target_modules=["q_proj", "v_proj","k_proj","o_proj"],
        lora_dropout=0.05,
        bias="none",
        task_type="CAUSAL_LM",
    )
    model = get_peft_model(model, lora_config).cuda()

    model.eval()
    return model, tokenizer, image_processor, image_token_len, mm_use_im_start_end
    
def load_biomedclip_model():
    """Loads the BiomedCLIP model and tokenizer."""
    biomedclip_model_name = 'microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224'
    processor = AutoProcessor.from_pretrained(biomedclip_model_name)
    model = AutoModel.from_pretrained(biomedclip_model_name).cuda().eval()
    return model, processor


class KeywordsStoppingCriteria(StoppingCriteria):
    """Custom stopping criteria for generation."""
    def __init__(self, keywords, tokenizer, input_ids):
        self.keywords = keywords
        self.tokenizer = tokenizer
        self.start_len = None
        self.input_ids = input_ids

    def __call__(self, output_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        if self.start_len is None:
            self.start_len = self.input_ids.shape[1]
        else:
            outputs = self.tokenizer.batch_decode(output_ids[:, self.start_len:], skip_special_tokens=True)[0]
            for keyword in self.keywords:
                if keyword in outputs:
                    return True
        return False

def compute_similarity(image, text, biomedclip_model, biomedclip_processor):
    """Computes similarity scores using BiomedCLIP."""
    with torch.no_grad():
        inputs = biomedclip_processor(text=text, images=image, return_tensors="pt", padding=True).to(biomedclip_model.device)
        outputs = biomedclip_model(**inputs)
        image_embeds = outputs.image_embeds
        text_embeds = outputs.text_embeds
    image_embeds = F.normalize(image_embeds, dim=-1)
    text_embeds = F.normalize(text_embeds, dim=-1)
    similarity = (text_embeds @ image_embeds.transpose(-1, -2)).squeeze()
    return similarity

@torch.no_grad()
def eval_llava_model(llava_model, llava_tokenizer, llava_image_processor, image, question, image_token_len, mm_use_im_start_end, max_new_tokens, temperature):
    """Evaluates the LLaVA model for a given image and question."""
    
    image_list = []
    image_tensor = llava_image_processor.preprocess(image, return_tensors='pt')['pixel_values'][0]  # 3, 224, 224
    image_list.append(image_tensor)
    image_idx = 1

    if mm_use_im_start_end:
        qs = DEFAULT_IM_START_TOKEN + DEFAULT_IMAGE_PATCH_TOKEN * image_token_len * image_idx + DEFAULT_IM_END_TOKEN + question
    else:
        qs = DEFAULT_IMAGE_PATCH_TOKEN * image_token_len * image_idx + '\n' + question

    conv = conv_templates["simple"].copy()
    conv.append_message(conv.roles[0], qs)
    prompt = conv.get_prompt()
    inputs = llava_tokenizer([prompt])

    image_tensor = torch.stack(image_list, dim=0).half().cuda()
    input_ids = torch.as_tensor(inputs.input_ids).cuda()

    keywords = ['###']
    stopping_criteria = KeywordsStoppingCriteria(keywords, llava_tokenizer, input_ids)

    with autocast():
        output_ids = llava_model.generate(
            input_ids=input_ids,
            images=image_tensor,
            do_sample=True,
            temperature=temperature,
            max_new_tokens=max_new_tokens,
            stopping_criteria=[stopping_criteria]
        )

    input_token_len = input_ids.shape[1]
    n_diff_input_output = (input_ids != output_ids[:, :input_token_len]).sum().item()
    if n_diff_input_output > 0:
        print(f'[Warning] Sample: {n_diff_input_output} output_ids are not the same as the input_ids')
    outputs = llava_tokenizer.batch_decode(output_ids[:, input_token_len:], skip_special_tokens=True)[0]

    while True:
        cur_len = len(outputs)
        outputs = outputs.strip()
        for pattern in ['###', 'Assistant:', 'Response:']:
            if outputs.startswith(pattern):
                outputs = outputs[len(pattern):].strip()
        if len(outputs) == cur_len:
            break

    try:
        index = outputs.index(conv.sep)
    except ValueError:
        outputs += conv.sep
        index = outputs.index(conv.sep)

    outputs = outputs[:index].strip()
    print(outputs)
    return outputs

#---------------------------------
#++++++++     Gradio     ++++++++++
#---------------------------------

SHARED_UI_WARNING = f'''### [NOTE] It is possible that you are waiting in a lengthy queue.
You can duplicate and use it with a paid private GPU.
<a class="duplicate-button" style="display:inline-block" target="_blank" href="https://huggingface.co/spaces/Vision-CAIR/minigpt4?duplicate=true"><img style="margin-top:0;margin-bottom:0" src="https://huggingface.co/datasets/huggingface/badges/raw/main/duplicate-this-space-xl-dark.svg" alt="Duplicate Space"></a>
Alternatively, you can also use the demo on our [project page](https://minigpt-4.github.io).
'''

def gradio_reset(chat_state, img_list):
    """Resets the chat state and image list."""
    if chat_state is not None:
        chat_state.messages = []
    if img_list is not None:
        img_list = []
    return None, gr.update(value=None, interactive=True), gr.update(placeholder='Please upload your medical image first', interactive=False), gr.update(value="Upload & Start Analysis", interactive=True), chat_state, img_list

def upload_img(gr_img, text_input, chat_state):
    """Handles image upload."""
    if gr_img is None:
        return None, None, gr.update(interactive=True), chat_state, None
    img_list = [gr_img]
    return gr.update(interactive=False), gr.update(interactive=True, placeholder='Type and press Enter'), gr.update(value="Start Analysis", interactive=False), chat_state, img_list


def gradio_ask(user_message, chatbot, chat_state):
    """Handles user input."""
    if not user_message:
        return gr.update(interactive=True, placeholder='Input should not be empty!'), chatbot, chat_state
    chatbot = chatbot + [[user_message, None]]
    return '', chatbot, chat_state

@spaces.GPU
def gradio_answer(chatbot, chat_state, img_list, llava_model, llava_tokenizer, llava_image_processor, image_token_len, mm_use_im_start_end, max_new_token, temperature, biomedclip_model, biomedclip_processor):
    """Generates and adds the bot's response to the chatbot using LLaVA"""
    if not img_list:
        return chatbot, chat_state, img_list
        
    # compute similarity using biomedclip
    similarity_score = compute_similarity(img_list[0],chatbot[-1][0], biomedclip_model, biomedclip_processor)
    print(f'Similarity Score is: {similarity_score}')
    
    # prepare the input for LLAVA
    llava_input_text = f"Based on the image and query provided the similarity score is {similarity_score:.3f}. " + chatbot[-1][0]
    llm_message = eval_llava_model(llava_model, llava_tokenizer, llava_image_processor, img_list[0], llava_input_text, image_token_len, mm_use_im_start_end, max_new_token, temperature)
    
    chatbot[-1][1] = llm_message
    return chatbot, chat_state, img_list

title = """<h1 align="center">Medical Image Analysis Tool</h1>"""
description = """<h3>Upload medical images, ask questions, and receive analysis.</h3>"""
examples_list=[
                    ["./case1.png", "Analyze the X-ray for any abnormalities."],
                    ["./case2.jpg", "What type of disease may be present?"],
                    ["./case1.png","What is the anatomical structure shown here?"]
                ]

# Load models and related resources outside of the Gradio block for loading on startup
llava_model, llava_tokenizer, llava_image_processor, image_token_len, mm_use_im_start_end = load_llava_model()
biomedclip_model, biomedclip_processor = load_biomedclip_model()


with gr.Blocks() as demo:
    gr.Markdown(title)
    # gr.Markdown(SHARED_UI_WARNING)
    gr.Markdown(description)

    with gr.Row():
        with gr.Column(scale=0.5):
            image = gr.Image(type="pil", label="Medical Image")
            upload_button = gr.Button(value="Upload & Start Analysis", interactive=True, variant="primary")
            clear = gr.Button("Restart")

            max_new_token = gr.Slider(
                minimum=1,
                maximum=512,
                value=128,
                step=1,
                interactive=True,
                label="Max new tokens"
            )

            temperature = gr.Slider(
                minimum=0.1,
                maximum=2.0,
                value=0.3,
                step=0.1,
                interactive=True,
                label="Temperature",
            )

        with gr.Column():
            chat_state = gr.State()
            img_list = gr.State()
            chatbot = gr.Chatbot(label='Medical Analysis')
            text_input = gr.Textbox(label='Analysis Query', placeholder='Please upload your medical image first', interactive=False)
            gr.Examples(examples=examples_list, inputs=[image, text_input])

    upload_button.click(upload_img, [image, text_input, chat_state], [image, text_input, upload_button, chat_state, img_list])

    text_input.submit(gradio_ask, [text_input, chatbot, chat_state], [text_input, chatbot, chat_state]).then(
        gradio_answer, [chatbot, chat_state, img_list, llava_model, llava_tokenizer, llava_image_processor, image_token_len, mm_use_im_start_end, max_new_token, temperature, biomedclip_model, biomedclip_processor], [chatbot, chat_state, img_list]
    )
    clear.click(gradio_reset, [chat_state, img_list], [chatbot, image, text_input, upload_button, chat_state, img_list], queue=False)


demo.launch()