app.py

import streamlit as st
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, AutoProcessor
from peft import PeftModel
from PIL import Image
import io

# Page configuration
st.set_page_config(
    page_title="ViTCM_LLM Tongue Diagnosis",
    page_icon="🖼️",
    layout="wide"
)

# Title
st.title("🖼️ ViTCM_LLM Tongue Diagnosis")
st.markdown("**ViTCM_LLM - Traditional Chinese Medicine Tongue Diagnosis Model**")

# Model loading
@st.cache_resource
def load_model():
    """Load the ViTCM_LLM model for TCM tongue diagnosis."""
    try:
        # Tokenizer and processor
        tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-VL-32B-Instruct")
        processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-32B-Instruct")
        
        # Base model
        base_model = AutoModelForCausalLM.from_pretrained(
            "Qwen/Qwen2.5-VL-32B-Instruct",
            torch_dtype=torch.float16,
            device_map="auto"
        )
        
        # LoRA adapter
        model = PeftModel.from_pretrained(base_model, "Mark-CHAE/shezhen")
        
        return model, tokenizer, processor
    except Exception as e:
        st.error(f"Model loading failed: {e}")
        return None, None, None

# Sidebar
with st.sidebar:
    st.header("⚙️ Settings")
    
    # Inference parameters
    max_length = st.slider("Max tokens", 100, 1024, 512)
    temperature = st.slider("Temperature", 0.1, 2.0, 0.7, 0.1)
    top_p = st.slider("Top-p", 0.1, 1.0, 0.9, 0.05)
    
    # Model load button
    if st.button("🚀 Load Model", type="primary"):
        with st.spinner("Loading ViTCM_LLM model..."):
            model, tokenizer, processor = load_model()
            if model is not None:
                st.session_state.model = model
                st.session_state.tokenizer = tokenizer
                st.session_state.processor = processor
                st.session_state.model_loaded = True
                st.success("✅ ViTCM_LLM model loaded successfully!")

# Main content
if not st.session_state.get('model_loaded', False):
    st.info("👈 Click 'Load Model' button in the sidebar to start tongue diagnosis.")
    st.stop()

# Image upload
st.header("📸 Tongue Image Upload")
uploaded_file = st.file_uploader(
    "Upload a tongue image for TCM diagnosis",
    type=['png', 'jpg', 'jpeg']
)

if uploaded_file is not None:
    # Display image
    image = Image.open(uploaded_file)
    st.image(image, caption="Uploaded tongue image", use_column_width=True)
    
    # Question input
    st.header("❓ Tongue Diagnosis Question")
    question = st.text_area(
        "Ask a question about the tongue image for TCM diagnosis",
        placeholder="e.g., 根据图片判断舌诊内容",
        height=100
    )
    
    # Analyze button
    if st.button("🔍 Analyze Tongue", type="primary") and question.strip():
        with st.spinner("Analyzing tongue for TCM diagnosis..."):
            try:
                # Construct prompt
                prompt = f"<|im_start|>user\n<image>\n{question}<|im_end|>\n<|im_start|>assistant\n"
                
                # Process inputs
                inputs = st.session_state.processor(
                    text=prompt,
                    images=image,
                    return_tensors="pt"
                )
                
                # Inference
                with torch.no_grad():
                    outputs = st.session_state.model.generate(
                        **inputs,
                        max_length=max_length,
                        temperature=temperature,
                        top_p=top_p,
                        do_sample=True,
                        pad_token_id=st.session_state.tokenizer.eos_token_id
                    )
                
                # Process results
                response = st.session_state.tokenizer.decode(outputs[0], skip_special_tokens=True)
                answer = response.split("<|im_start|>assistant")[-1].strip()
                
                # Display results
                st.header("💡 TCM Tongue Diagnosis")
                st.markdown(f"**Question:** {question}")
                st.markdown(f"**Diagnosis:** {answer}")
                
            except Exception as e:
                st.error(f"Error occurred during tongue analysis: {e}")

# Usage examples
with st.expander("📚 Tongue Diagnosis Examples"):
    st.markdown("""
    ### Tongue Diagnosis Questions:
    - 根据图片判断舌诊内容
    - 分析舌头的颜色和形状
    - 判断舌苔的厚薄和颜色
    - 分析舌头的裂纹和斑点
    - 评估舌头的整体健康状况
    """)

# Model information
with st.expander("ℹ️ Model Information"):
    st.markdown("""
    ### ViTCM_LLM - Traditional Chinese Medicine Tongue Diagnosis Model
    
    - **Base Model**: Qwen/Qwen2.5-VL-32B-Instruct
    - **Adapter**: Mark-CHAE/shezhen (ViTCM_LLM)
    - **Language**: Chinese
    - **License**: Apache-2.0
    - **Specialization**: Traditional Chinese Medicine Tongue Diagnosis
    """)

st.markdown("---")
st.markdown("**ViTCM_LLM Tongue Diagnosis** | Powered by Qwen2.5-VL-32B-Instruct")