app.py

from fix_int8 import fix_pytorch_int8
fix_pytorch_int8()


# Credit:
# https://huggingface.co/spaces/ljsabc/Fujisaki/blob/main/app.py


import torch
import gradio as gr
from threading import Thread
from model import model, tokenizer
from session import db, logger, log_sys_info
from transformers import AutoTokenizer, GenerationConfig, AutoModel


max_length = 224
default_start = ["你是Kuma，请和我聊天，每句话以两个竖杠分隔。", "好的，你想聊什么？"]

gr_title = """<h1 align="center">KumaGLM</h1>
<h3 align='center'>这是一个 AI Kuma，你可以与他聊天，或者直接在文本框按下Enter</h3>
<p align='center'>采样范围 2020/06/13 - 2023/04/15</p>
<p align='center'>GitHub Repo: <a class="github-button" href="https://github.com/KumaTea/ChatGLM" aria-label="Star KumaTea/ChatGLM on GitHub">KumaTea/ChatGLM</a></p>
<script async defer src="https://buttons.github.io/buttons.js"></script>
"""
gr_footer =  """<p align='center'>
本项目基于
<a href='https://github.com/ljsabc/Fujisaki' target='_blank'>ljsabc/Fujisaki</a>
，模型采用
<a href='https://huggingface.co/THUDM/chatglm-6b' target='_blank'>THUDM/chatglm-6b</a>
。
</p>
<p align='center'>
<em>每天起床第一句！</em>
</p>"""


def evaluate(context, temperature, top_p):
    generation_config = GenerationConfig(
        temperature=temperature,
        top_p=top_p,
        # top_k=top_k,
        #repetition_penalty=1.1,
        num_beams=1,
        do_sample=True,
    )
    with torch.no_grad():
        # input_text = f"Context: {context}Answer: " 
        # input_text = '||'.join(default_start) + '||'
        # No need for starting prompt in API
        if not context.endswith('||'):
            context += '||'
        # logger.info('[API] Request: ' + context)
        ids = tokenizer([context], return_tensors="pt")
        inputs = ids.to("cpu")
        out = model.generate(
            **inputs,
            max_length=max_length,
            generation_config=generation_config
        )
        out = out.tolist()[0]
        decoder_output = tokenizer.decode(out)
        # out_text = decoder_output.split("Answer: ")[1]
        out_text = decoder_output
        logger.info('[API] Results: ' + out_text.replace('\n', '<br>'))
        return out_text


def evaluate_wrapper(context, temperature, top_p):
    db.lock()
    index = db.index
    db.set(index, prompt=context)
    result = evaluate(context, temperature, top_p)
    db.set(index, result=result)
    db.unlock()
    return result


def api_wrapper(context='', temperature=0.5, top_p=0.8, query=0):
    query = int(query)
    assert context or query
    
    return_json = {
        'status': '',
        'code': 0,
        'message': '',
        'index': 0,
        'result': ''
    }

    if context:
        if db.islocked():
            logger.info(f'[API] Request: {context}, Status: busy')
            return_json['status'] = 'busy'
            return_json['code'] = 503
            return_json['message'] = '[context] Server is busy, please try again later.'
            return return_json
        else:
            for index in db.prompts:
                if db.prompts[index] == context:
                    return_json['status'] = 'done'
                    return_json['code'] = 200
                    return_json['message'] = '[context] Request cached.'
                    return_json['index'] = index
                    return_json['result'] = db.results[index]
                    return return_json
            # new
            index = db.index
            t = Thread(target=evaluate_wrapper, args=(context, temperature, top_p))
            t.start()
            logger.info(f'[API] Request: {context}, Status: processing, Index: {index}')
            return_json['status'] = 'processing'
            return_json['code'] = 202
            return_json['message'] = '[context] Request accepted, please check back later.'
            return_json['index'] = index
            return return_json
    else:  # query
        if query in db.prompts and query in db.results:
            logger.info(f'[API] Query: {query}, Status: hit')
            return_json['status'] = 'done'
            return_json['code'] = 200
            return_json['message'] = '[query] Request processed.'
            return_json['index'] = query
            return_json['result'] = db.results[query]
            return return_json
        else:
            if db.islocked():
                logger.info(f'[API] Query: {query}, Status: processing')
                return_json['status'] = 'processing'
                return_json['code'] = 202
                return_json['message'] = '[query] Request in processing, please check back later.'
                return_json['index'] = query
                return return_json
            else:
                logger.info(f'[API] Query: {query}, Status: error')
                return_json['status'] = 'error'
                return_json['code'] = 404
                return_json['message'] = '[query] Index not found.'
                return_json['index'] = query
                return return_json


def evaluate_stream(msg, history, temperature, top_p):
    generation_config = GenerationConfig(
        temperature=temperature,
        top_p=top_p,
        #repetition_penalty=1.1,
        num_beams=1,
        do_sample=True,
    )
    if not msg:
        msg = '……'

    history.append([msg, ""])

    context = '||'.join(default_start) + '||'
    if len(history) > 4:
        history.pop(0)

    for j in range(len(history)):
        history[j][0] = history[j][0].replace("<br>", "")

    # concatenate context
    for h in history[:-1]:
        context += h[0] + "||" + h[1] + "||"

    context += history[-1][0] + "||"
    context = context.replace(r'<br>', '')

    # TODO: Avoid the tokens are too long.
    # CUTOFF = 224
    while len(tokenizer.encode(context)) > max_length:
        # save 15 token size for the answer
        context = context[15:]

    h = []
    logger.info('[UI] Request: ' + context)
    for response, h in model.stream_chat(tokenizer, context, h, max_length=max_length, top_p=top_p, temperature=temperature):
        history[-1][1] = response
        yield history, ""
    logger.info('[UI] Results: ' + response.replace('\n', '<br>'))


with gr.Blocks() as demo:
    gr.HTML(gr_title)
    # state = gr.State()
    with gr.Row():
        with gr.Column(scale=2):
            temp = gr.components.Slider(minimum=0, maximum=1.1, value=0.5, label="Temperature",
                info="温度参数，越高的温度生成的内容越丰富，但是有可能出现语法问题。小的温度也能帮助生成更相关的回答。")
            top_p = gr.components.Slider(minimum=0.5, maximum=1.0, value=0.8, label="Top-p",
                info="top-p参数，只输出前p>top-p的文字，越大生成的内容越丰富，但也可能出现语法问题。数字越小似乎上下文的衔接性越好。")
            #code = gr.Textbox(label="temp_output", info="解码器输出")
            #top_k = gr.components.Slider(minimum=1, maximum=200, step=1, value=25, label="Top k",
            #    info="top-k参数，下一个输出的文字会从top-k个文字中进行选择，越大生成的内容越丰富，但也可能出现语法问题。数字越小似乎上下文的衔接性越好。")
            
        with gr.Column(scale=3):
            chatbot = gr.Chatbot(label="聊天框", info="")
            msg = gr.Textbox(label="输入框", placeholder="最近过得怎么样？",
                info="输入你的内容，按 [Enter] 发送。什么都不填经常会出错。")
            clear = gr.Button("清除聊天")

        api_handler = gr.Button("API", visible=False)
        api_index = gr.Number(visible=False)
        api_result = gr.JSON(visible=False)
        info_handler = gr.Button("Info", visible=False)
        info_text = gr.Textbox('System info logged. Check it in the log viewer.', visible=False)


    msg.submit(evaluate_stream, [msg, chatbot, temp, top_p], [chatbot, msg])
    clear.click(lambda: None, None, chatbot, queue=False)
    api_handler.click(api_wrapper, [msg, temp, top_p, api_index], api_result, api_name='chat')
    info_handler.click(log_sys_info, None, info_text, api_name='info')
    gr.HTML(gr_footer)

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