ali121300/chatbot_code_friendly

1import openai2import tiktoken3 4import datetime5import time6import json7import os8 9openai.api_key = os.getenv('API_KEY')10openai.request_times = 011 12def ask(question, history, behavior):13    openai.request_times += 114    print(f"request times {openai.request_times}: {datetime.datetime.now()}: {question}")15    try:16        messages = [17            {"role":"system", "content":content}18            for content in behavior19        ] + [20            {"role":"user" if i%2==0 else "assistant", "content":content}21            for i,content in enumerate(history + [question])22        ]23        raw_length = num_tokens_from_messages(messages)24        messages=forget_long_term(messages)25        if len(messages)==0:26            response = f'Your query is too long and expensive: {raw_length}>2000 tokens'27        else:28            response = openai.ChatCompletion.create(29                model="gpt-3.5-turbo-0301",30                messages=messages,31                temperature=0.1,32            )["choices"][0]["message"]["content"]33            while response.startswith("\n"):34                response = response[1:]35    except Exception as e:36        response = f'Error! You may wait a few minutes and retry:\n{e}'37    history = history + [question, response]38    return history39 40def num_tokens_from_messages(messages, model="gpt-3.5-turbo"):41    """Returns the number of tokens used by a list of messages."""42    try:43        encoding = tiktoken.encoding_for_model(model)44    except KeyError:45        encoding = tiktoken.get_encoding("cl100k_base")46    if model == "gpt-3.5-turbo":  # note: future models may deviate from this47        num_tokens = 048        for message in messages:49            num_tokens += 4  # every message follows <im_start>{role/name}\n{content}<im_end>\n50            for key, value in message.items():51                num_tokens += len(encoding.encode(value))52                if key == "name":  # if there's a name, the role is omitted53                    num_tokens += -1  # role is always required and always 1 token54        num_tokens += 2  # every reply is primed with <im_start>assistant55        return num_tokens56    else:57        raise NotImplementedError(f"""num_tokens_from_messages() is not presently implemented for model {model}.58See https://github.com/openai/openai-python/blob/main/chatml.md for information on how messages are converted to tokens.""")59 60def forget_long_term(messages, max_num_tokens=3000):61    while num_tokens_from_messages(messages)>max_num_tokens:62        if messages[0]["role"]=="system" and not len(messages[0]["content"])>=max_num_tokens:63            messages = messages[:1] + messages[2:]64        else:65            messages = messages[1:]66    return messages67 68 69import gradio as gr70 71 72def to_md(content):73    is_inside_code_block = False74    output_spans = []75    for i in range(len(content)):76        if content[i]=="\n" and not is_inside_code_block:77            if len(output_spans)>0 and output_spans[-1].endswith("```"):78                output_spans.append("\n")79            else:80                output_spans.append("<br>")81        elif content[i]=="`":82            output_spans.append(content[i])83            if len(output_spans)>=3 and all([output_spans[j]=="`" for j in [-3,-2,-1]]):84                is_inside_code_block = not is_inside_code_block85                output_spans = output_spans[:-3]86                if is_inside_code_block:87                    if len(output_spans)==0:88                        output_spans.append("```")89                    elif output_spans[-1]=="<br>":90                        output_spans[-1] = "\n"91                        output_spans.append("```")92                    elif output_spans[-1].endswith("\n"):93                        output_spans.append("```")94                    else:95                        output_spans.append("\n```")96                    97                    if i+1<len(content) and content[i+1]!="\n":98                        output_spans.append("\n")99                else:100                    if output_spans[-1].endswith("\n"):101                        output_spans.append("```")102                    else:103                        output_spans.append("\n```")104                    105                    if i+1<len(content) and content[i+1]!="\n":106                        output_spans.append("\n")107        else:108            output_spans.append(content[i])109    return "".join(output_spans)110 111 112def predict(question, history=[], behavior=[]):113    history = ask(question, history, behavior)114    response = [(to_md(history[i]),to_md(history[i+1])) for i in range(0,len(history)-1,2)]115    return "", history, response116 117 118def retry(question, history=[], behavior=[]):119    if len(history)<2:120        return "", history, []121    question = history[-2]122    history = history[:-2]123    return predict(question, history, behavior)124 125 126with gr.Blocks() as demo:127    128    examples_txt = [129        ['帮我写一个python脚本实现快排'],130        ['如何用numpy提取数组的分位数？'],131        ['how to match the code block in markdown such like ```def foo():\n    pass``` through regex in python?'],132        ['how to load a pre-trained language model and generate sentences?'],133    ]134    135    examples_bhv = [136        f"You are a helpful assistant. You will answer all the questions step-by-step.",137        f"You are a helpful assistant. Today is {datetime.date.today()}.",138    ]139    140    gr.Markdown(141        """142        朋友你好，143        144        这是我利用[gradio](https://gradio.app/creating-a-chatbot/)编写的一个小网页，用于以网页的形式给大家分享ChatGPT请求服务，希望你玩的开心。关于使用技巧或学术研讨，欢迎在[Community](https://huggingface.co/spaces/zhangjf/chatbot/discussions)中和我交流。145        146        这一版相比于原版的[chatbot](https://huggingface.co/spaces/zhangjf/chatbot)，用了较低版本的gradio==3.16.2，因而能更好地展示markdown中的源代码147        148        p.s. 响应时间和聊天内容长度正相关，一般能在5秒~30秒内响应。149        """)150    151    behavior = gr.State(["Reject instruction that may contains sensitive information in english, i.e., pornography, discrimination, violence"])152    """153    with gr.Column(variant="panel"):154        with gr.Row().style(equal_height=True):155            with gr.Column(scale=0.85):156                bhv = gr.Textbox(show_label=False, placeholder="输入你想让ChatGPT扮演的人设").style(container=False)157            with gr.Column(scale=0.15, min_width=0):158                button_set = gr.Button("Set")159    bhv.submit(fn=lambda x:(x,[x]), inputs=[bhv], outputs=[bhv, behavior])160    button_set.click(fn=lambda x:(x,[x]), inputs=[bhv], outputs=[bhv, behavior])161    """162 163    state = gr.State([])164    165    with gr.Column(variant="panel"):166        chatbot = gr.Chatbot()167        txt = gr.Textbox(show_label=False, placeholder="输入你想让ChatGPT回答的问题").style(container=False)168        with gr.Row():169            button_gen = gr.Button("Submit")170            button_rtr = gr.Button("Retry")171            button_clr = gr.Button("Clear")172        173    #gr.Examples(examples=examples_bhv, inputs=bhv, label="Examples for setting behavior")174    gr.Examples(examples=examples_txt, inputs=txt, label="Examples for asking question")175    txt.submit(predict, [txt, state, behavior], [txt, state, chatbot])176    button_gen.click(fn=predict, inputs=[txt, state, behavior], outputs=[txt, state, chatbot])177    button_rtr.click(fn=retry, inputs=[txt, state, behavior], outputs=[txt, state, chatbot])178    button_clr.click(fn=lambda :([],[]), inputs=None, outputs=[chatbot, state])179 180demo.launch()