import gradio as gr
from huggingface_hub import InferenceClient
from optimum.intel import OVModelForCausalLM
from transformers import AutoTokenizer, pipeline

# 載入模型和標記器
model_id = "hsuwill000/SmolLM2-135M-openvino"
model = OVModelForCausalLM.from_pretrained(model_id, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_id)

# 建立生成管道
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

def respond(message, history):
    # 將當前訊息與歷史訊息合併
    input_text = message if not history else history[-1]["content"] + " " + message
    input_text = message
    # 獲取模型的回應
    response = pipe(input_text, max_length=500, truncation=True, num_return_sequences=1)
    reply = response[0]['generated_text']
    
    # 返回新的消息格式
    print(f"Message: {message}")
    print(f"Reply: {reply}")
    return reply
    
# 設定 Gradio 的聊天界面
demo = gr.ChatInterface(fn=respond, title="Chat with Qwen(通義千問) 2.5-0.5B", description="與 Qwen2.5-0.5B-Instruct-openvino 聊天！", type='messages')

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