import os
import torch
import gradio as gr
import tempfile
import secrets
from pathlib import Path
from transformers import AutoModelForCausalLM, AutoTokenizer, BlipForConditionalGeneration, AutoProcessor
from PIL import Image

# Load Vision-Language Model
vl_model_name = "Salesforce/blip-image-captioning-large"
vl_model = BlipForConditionalGeneration.from_pretrained(vl_model_name)
vl_processor = AutoProcessor.from_pretrained(vl_model_name)

# Load Text Model
model_name = "Qwen/Qwen2.5-Math-1.5B-Instruct"
device = "cuda" if torch.cuda.is_available() else "cpu"
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name)

math_messages = []

def process_image(image, shouldConvert=False):
    global math_messages
    math_messages = []  # Reset when uploading an image

    if shouldConvert:
        new_img = Image.new('RGB', size=(image.width, image.height), color=(255, 255, 255))
        new_img.paste(image, (0, 0), mask=image)
        image = new_img

    # Convert the image to tensor
    inputs = vl_processor(images=image, return_tensors="pt")
    output = vl_model.generate(**inputs)
    description = vl_processor.batch_decode(output, skip_special_tokens=True)[0]

    return f"Math-related content detected: {description}"

def get_math_response(image_description, user_question):
    global math_messages
    if not math_messages:
        math_messages.append({'role': 'system', 'content': 'You are a helpful math assistant.'})
    math_messages = math_messages[:1]
    content = f'Image description: {image_description}\n\n' if image_description else ''
    query = f"{content}User question: {user_question}"
    math_messages.append({'role': 'user', 'content': query})
    model_inputs = tokenizer(query, return_tensors="pt").to(device)
    output = model.generate(**model_inputs, max_new_tokens=512)
    answer = tokenizer.decode(output[0], skip_special_tokens=True)
    yield answer.replace("\\", "\\\\")
    math_messages.append({'role': 'assistant', 'content': answer})

def math_chat_bot(image, sketchpad, question, state):
    current_tab_index = state["tab_index"]
    image_description = None
    if current_tab_index == 0:
        if image is not None:
            image_description = process_image(image)
    elif current_tab_index == 1:
        if sketchpad and sketchpad["composite"]:
            image_description = process_image(sketchpad["composite"], True)
    yield from get_math_response(image_description, question)

css = """
#qwen-md .katex-display { display: inline; }
#qwen-md .katex-display>.katex { display: inline; }
#qwen-md .katex-display>.katex>.katex-html { display: inline; }
"""

def tabs_select(e: gr.SelectData, _state):
    _state["tab_index"] = e.index


# 创建Gradio接口
with gr.Blocks(css=css) as demo:
    gr.HTML("""\
<p align="center"><img src="https://modelscope.oss-cn-beijing.aliyuncs.com/resource/qwen.png" style="height: 60px"/><p>"""
            """<center><font size=8>📖 Qwen2-Math Demo</center>"""
            """\
<center><font size=3>This WebUI is based on Qwen2-VL for OCR and Qwen2-Math for mathematical reasoning. You can input either images or texts of mathematical or arithmetic problems.</center>"""
            )
    state = gr.State({"tab_index": 0})
    with gr.Row():
        with gr.Column():
            with gr.Tabs() as input_tabs:
                with gr.Tab("Upload"):
                    input_image = gr.Image(type="pil", label="Upload"),
                with gr.Tab("Sketch"):
                    input_sketchpad = gr.Sketchpad(type="pil", label="Sketch", layers=False)
            input_tabs.select(fn=tabs_select, inputs=[state])
            input_text = gr.Textbox(label="input your question")
            with gr.Row():
                with gr.Column():
                    clear_btn = gr.ClearButton(
                        [*input_image, input_sketchpad, input_text])
                with gr.Column():
                    submit_btn = gr.Button("Submit", variant="primary")
        with gr.Column():
            output_md = gr.Markdown(label="answer",
                                    latex_delimiters=[{
                                        "left": "\\(",
                                        "right": "\\)",
                                        "display": True
                                    }, {
                                        "left": "\\begin\{equation\}",
                                        "right": "\\end\{equation\}",
                                        "display": True
                                    }, {
                                        "left": "\\begin\{align\}",
                                        "right": "\\end\{align\}",
                                        "display": True
                                    }, {
                                        "left": "\\begin\{alignat\}",
                                        "right": "\\end\{alignat\}",
                                        "display": True
                                    }, {
                                        "left": "\\begin\{gather\}",
                                        "right": "\\end\{gather\}",
                                        "display": True
                                    }, {
                                        "left": "\\begin\{CD\}",
                                        "right": "\\end\{CD\}",
                                        "display": True
                                    }, {
                                        "left": "\\[",
                                        "right": "\\]",
                                        "display": True
                                    }],
                                    elem_id="qwen-md")
        submit_btn.click(
            fn=math_chat_bot,
            inputs=[*input_image, input_sketchpad, input_text, state],
            outputs=output_md)
        
if __name__ == "__main__":
    demo.launch()