app.py

import gradio as gr
import os
import json
import requests
from langchain import FAISS
from langchain.embeddings import CohereEmbeddings, OpenAIEmbeddings
from langchain import VectorDBQA
from langchain.chat_models import ChatOpenAI
from prompts import MyTemplate
from langchain.prompts.chat import (
    ChatPromptTemplate,
    SystemMessagePromptTemplate,
    HumanMessagePromptTemplate,
)

# Streaming endpoint
API_URL = "https://api.openai.com/v1/chat/completions"
cohere_key = '5IRbILAbjTI0VcqTsktBfKsr13Lych9iBAFbLpkj'
faiss_store = './indexer/{}'
docsearch = None


def gen_conversation(conversations):
    messages = []
    for data in conversations:
        temp1 = {}
        temp1["role"] = "user"
        temp1["content"] = data[0]
        temp2 = {}
        temp2["role"] = "assistant"
        temp2["content"] = data[1]
        messages.append(temp1)
        messages.append(temp2)
    return messages


def predict(inputs, top_p, temperature, openai_api_key, enable_index, max_tokens, topic,
            chat_counter, chatbot=[], history=[]):
    global docsearch

    topic = topic[0]
    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {openai_api_key}"
    }

    print(f"chat_counter - {chat_counter}")
    print(f'Histroy - {history}')  # History: Original Input and Output in flatten list
    print(f'chatbot - {chatbot}')  # Chat Bot： 上一轮回复的[[user, AI]]

    history.append(inputs)
    # Debugging
    if enable_index:
        # Faiss 检索最近的embedding
        store = faiss_store.format(topic)
        if docsearch is None:
            print('Loading FAISS')
            docsearch = FAISS.load_local(store, OpenAIEmbeddings(openai_api_key=openai_api_key))
        else:
            print('Faiss already loaded')
        # 构建模板
        llm = ChatOpenAI(openai_api_key=openai_api_key, max_tokens=max_tokens)
        messages_combine = [
            SystemMessagePromptTemplate.from_template(MyTemplate['chat_combine_template']),
            HumanMessagePromptTemplate.from_template("{question}")
        ]
        p_chat_combine = ChatPromptTemplate.from_messages(messages_combine)
        messages_reduce = [
            SystemMessagePromptTemplate.from_template(MyTemplate['chat_reduce_template']),
            HumanMessagePromptTemplate.from_template("{question}")
        ]
        p_chat_reduce = ChatPromptTemplate.from_messages(messages_reduce)
        chain = VectorDBQA.from_chain_type(llm=llm, chain_type="map_reduce", vectorstore=docsearch,
                                           k=4,
                                           chain_type_kwargs={"question_prompt": p_chat_reduce,
                                                              "combine_prompt": p_chat_combine}
                                           )
        result = chain({"query": inputs})
        print(result)
        result = result['result']
        # 生成返回值
        history.append(result)
        chat = [(history[i], history[i + 1]) for i in range(0, len(history) - 1, 2)]
        chat_counter += 1
        yield chat, history, chat_counter
    else:
        if chat_counter == 0:
            messages = [{"role": "user", "content": f"{inputs}"}]
        else:
            # 如果有历史对话，把对话拼接进入上下文
            messages = gen_conversation(chatbot)
            messages.append({'role': 'user', 'content': inputs})
        # messages
        payload = {
            "model": "gpt-3.5-turbo",
            "messages": messages,  # [{"role": "user", "content": f"{inputs}"}],
            "temperature": temperature,  # 1.0,
            "top_p": top_p,  # 1.0,
            "n": 1,
            "stream": True,
            "presence_penalty": 0,
            "frequency_penalty": 0,
        }
        print(f"payload is - {payload}")

        chat_counter += 1

        # 请求OpenAI
        response = requests.post(API_URL, headers=headers, json=payload, stream=True)
        token_counter = 0
        partial_words = ""

        # 逐字返回
        counter = 0
        for chunk in response.iter_lines():
            if counter == 0:
                counter += 1
                continue
            counter += 1
            # check whether each line is non-empty
            if chunk:
                # decode each line as response data is in bytes
                delta = json.loads(chunk.decode()[6:])['choices'][0]["delta"]
                if len(delta) == 0:
                    break
                partial_words += delta["content"]
                # Keep Updating history
                if token_counter == 0:
                    history.append(" " + partial_words)
                else:
                    history[-1] = partial_words
                chat = [(history[i], history[i + 1]) for i in
                        range(0, len(history) - 1, 2)]  # convert to tuples of list
                token_counter += 1
                yield chat, history, chat_counter


def reset_textbox():
    return gr.update(value='')


with gr.Blocks(css="""#col_container {width: 1000px; margin-left: auto; margin-right: auto;}
                #chatbot {height: 520px; overflow: auto;}""") as demo:
    gr.HTML("""<h1 align="center">🚀Finance ChatBot🚀</h1>""")
    with gr.Column(elem_id="col_container"):
        openai_api_key = gr.Textbox(type='password', label="输入OPEN API Key")

        # inputs, top_p, temperature, top_k, repetition_penalty
        with gr.Accordion("Parameters", open=True):
            with gr.Row():
                top_p = gr.Slider(minimum=-0, maximum=1.0, value=0.9, step=0.05, interactive=True,
                                  label="Top-p (nucleus sampling)", )
                temperature = gr.Slider(minimum=-0, maximum=5.0, value=0.8, step=0.1, interactive=True,
                                        label="Temperature", )
                max_tokens = gr.Slider(minimum=100, maximum=1000, value=200, step=100, interactive=True,
                                       label="Max Tokens", )
                chat_counter = gr.Number(value=0, precision=0, label='对话轮数')

            with gr.Row():
                enable_index = gr.Checkbox(label='是', info='开启文档问答模式/聊天模式')
                enable_search = gr.Checkbox(label='是', info='是否使用搜索')
                topic = gr.CheckboxGroup(["两会", "数字经济", "硅谷银行"], label='使用文档索引')

        chatbot = gr.Chatbot(elem_id='chatbot')
        inputs = gr.Textbox(placeholder="您有什么问题可以问我", label="输入数字经济，两会，硅谷银行相关的提问")
        state = gr.State([])

        with gr.Row():
            clear = gr.Button("Clear Conversation")
            run = gr.Button("Run")

    inputs.submit(predict,
                  [inputs, top_p, temperature, openai_api_key, enable_index, max_tokens, topic, chat_counter, chatbot,
                   state],
                  [chatbot, state, chat_counter], )
    run.click(predict,
              [inputs, top_p, temperature, openai_api_key, enable_index, max_tokens, topic, chat_counter, chatbot,
               state],
              [chatbot, state, chat_counter], )

    # 每次对话结束都重置对话
    clear.click(reset_textbox, [], [inputs], queue=False)
    inputs.submit(reset_textbox, [], [inputs])

    demo.queue().launch(debug=True)