app.py

# app.py
import torch
import gradio as gr
from transformers import AutoModelForCausalLM, BitsAndBytesConfig
from deepseek_vl.models import VLChatProcessor, MultiModalityCausalLM

# ==== 模型設定 ====
model_path = "deepseek-ai/deepseek-vl-7b-chat"

# BitsAndBytes 4-bit 量化設定
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_use_double_quant=True
)

# 載入 processor 和 tokenizer
vl_chat_processor: VLChatProcessor = VLChatProcessor.from_pretrained(model_path)
tokenizer = vl_chat_processor.tokenizer

# 載入模型
vl_gpt: MultiModalityCausalLM = AutoModelForCausalLM.from_pretrained(
    model_path,
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True
).eval()

# ==== 對話歷史 ====
chat_history = []

# ==== 文字+圖片推理函式 ====
def chat_with_image(image, user_message):
    global chat_history
    try:
        # 建立對話內容
        conversation = chat_history.copy()
        conversation.append({
            "role": "User",
            "content": "<image_placeholder>" + user_message,
            "images": [image] if image else []
        })
        conversation.append({"role": "Assistant", "content": ""})

        # 準備輸入
        prepare_inputs = vl_chat_processor(
            conversations=conversation,
            images=[image] if image else [],
            force_batchify=True
        ).to(vl_gpt.device)

        # 轉成 dict，並正確處理 dtype
        prepare_inputs = {k: getattr(prepare_inputs, k) for k in prepare_inputs.__dataclass_fields__.keys()}
        new_inputs = {}
        for k, v in prepare_inputs.items():
            if torch.is_tensor(v):
                if k in ["input_ids", "labels"]:
                    new_inputs[k] = v.to(torch.long)
                else:
                    new_inputs[k] = v.to(torch.float16)
            else:
                new_inputs[k] = v
        prepare_inputs = new_inputs

        # 取得 embeddings
        inputs_embeds = vl_gpt.prepare_inputs_embeds(**prepare_inputs)

        # 生成回答
        outputs = vl_gpt.language_model.generate(
            inputs_embeds=inputs_embeds,
            attention_mask=prepare_inputs["attention_mask"],
            pad_token_id=tokenizer.eos_token_id,
            bos_token_id=tokenizer.bos_token_id,
            eos_token_id=tokenizer.eos_token_id,
            max_new_tokens=128,
            do_sample=False,
            use_cache=True
        )

        # 解碼
        answer = tokenizer.decode(outputs[0].cpu().tolist(), skip_special_tokens=True)

        # 更新歷史
        chat_history.append((user_message, answer))
        return answer, chat_history

    except Exception as e:
        return f"Error: {str(e)}", chat_history

def reset_chat():
    global chat_history
    chat_history = []
    return "", []

# ==== Gradio Web UI ====
with gr.Blocks() as demo:
    gr.Markdown("# DeepSeek-VL-7B-Chat Demo (4-bit, float16)")
    with gr.Row():
        image_input = gr.Image(type="pil", label="Upload Image")
        text_input = gr.Textbox(lines=2, placeholder="Ask about the image...")
    with gr.Row():
        submit_btn = gr.Button("Submit")
        reset_btn = gr.Button("Reset Chat")
    output_text = gr.Textbox(label="Answer")
    chat_display = gr.Chatbot(label="Chat History")

    submit_btn.click(chat_with_image, inputs=[image_input, text_input], outputs=[output_text, chat_display])
    reset_btn.click(reset_chat, inputs=[], outputs=[output_text, chat_display])

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