import gradio as gr
from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained("silver/chatglm-6b-slim", trust_remote_code=True)
model = AutoModel.from_pretrained("silver/chatglm-6b-slim", trust_remote_code=True).half().cuda()

def greet(name):
    response, history = model.chat(tokenizer, "你好", history=[])
    #return "Hello " + name + "!!"
    return response
    

iface = gr.Interface(fn=greet, inputs="text", outputs="text")
iface.launch()