app.py

# -*- encoding: utf-8 -*-
import gradio as gr
import random
import time
from concurrent.futures import ThreadPoolExecutor
import asyncio
from functools import partial
from volcengine.maas import MaasService, MaasException, ChatRole
maas = MaasService('maas-api.ml-platform-cn-beijing.volces.com', 'cn-beijing', connection_timeout=24000, socket_timeout=24000)
maas.set_ak("AKLTY2ZjOWM0NmFlNTEwNDBhM2EyYTg4OTgzYTUyYTc2NjU")

maas.set_sk("T0Rjd01HUTFaVEF3Tm1FME5EWXhNRGhpTURreE1ETTROalkwTnpreU1UVQ==")

def llm_infer(endpoint_id, messages, agent_network_output=True):
    req = {
        "model":{
           "endpoint_id":endpoint_id
        },
        "parameters": {
            "max_prompt_tokens": 3800,  # 最大prompt，自动截断前面的输入。0-不生效
            "max_new_tokens": 0,  # 输出文本的最大tokens限制。0-不生效
            "min_new_tokens": 0,  # 输出文本的最小tokens限制。0-不生效
            "temperature":1,  # 用于控制生成文本的随机性和创造性，Temperature值越大随机性越大。取值范围0~1
            "top_p": 0.7,  # 用于控制输出tokens的多样性，TopP值越大输出的tokens类型越丰富。取值范围0~1
            "top_k": 0,  # 选择预测值最大的k个token进行采样，取值范围0-1000。0-不生效
        },
            "messages": messages,
        }
    while True:
        try:
            resp = maas.chat(req)
            break
        except MaasException as e:

            print(e)
            continue

    return resp['choice']['message']['content']

def respond(message, history, ep: gr.TextArea, sp:gr.TextArea):
    messages = [
        {
        'role':'system',
        'content':sp
        },
        ]
    for part in history:
        messages.extend(
            [
                {
                    'role':'user',
                    'content':part[0]
                },
                {
                    'role':'assistant',
                    'content':part[1]
                }
            ]
        )
    messages.append({
        'role':'user',
        'content':message
    })
    print(messages)
    ret=llm_infer(ep.strip(), messages)
    history.append([
        message,
        ret
    ])
    #if len(history) % 2 == 0:
    return "", history


with gr.Blocks(fill_height=True).queue(default_concurrency_limit=10) as demo:
    with gr.Row(): 
        with gr.Column():
            chatbot = gr.Chatbot()
            chatbot.height = 500
        with gr.Column():
            gr.Label("input sp/ep of bot") # 增加些描述
            sp=gr.TextArea(label="你是...", lines=1, max_lines=5,value='')
            ep=gr.TextArea(label='ep1', lines=1, max_lines=1, value='')
            
            
    with gr.Row():
        chat_input = gr.Textbox(label="input")
    
    def handle_click(chatbot, ep, sp, evt:gr.SelectData):
        messages = [
            {
            'role':'system',
            'content':sp
            },
        ]
        index = evt.index[0]
        print(index)
        for part in chatbot[:index]:
            messages.extend(
                [
                    {
                        'role':'user',
                        'content':part[0]
                    },
                    {
                        'role':'assistant',
                        'content':part[1]
                    }
                ]
            )
        ret = llm_infer(ep, messages)
        print(ret)
        print(chatbot[:index] + [[chatbot[index][0], ret]] + chatbot[index + 1:])
        return chatbot[:index] + [[chatbot[index][0], ret]] + chatbot[index + 1:]
   
    chat_input.submit(respond, [chat_input,chatbot, ep, sp ],[chat_input, chatbot])                    
    #chatbot.select(handle_click, [chatbot, ep, sp],[chatbot],show_progress='hidden')
    clear = gr.ClearButton([chat_input, chatbot],)
    
    
    
    
demo.launch(share=True)