stage llm model interface

feat: add function to parse Golang projects

README.md

@@ -36,14 +36,16 @@ https://github.com/polarwinkel/mdtex2html
 自定义快捷键 | 支持自定义快捷键
 配置代理服务器 | 支持配置代理服务器
 模块化设计 | 支持自定义高阶的实验性功能
-自我程序剖析 | [实验性功能] 一键读懂本项目的源代码
-程序剖析 | [实验性功能] 一键可以剖析其他Python/C++项目
-读论文 | [实验性功能] 一键解读latex论文全文并生成摘要
-批量注释生成 | [实验性功能] 一键批量生成函数注释
-chat分析报告生成 | [实验性功能] 运行后自动生成总结汇报
+自我程序剖析 | [函数插件] 一键读懂本项目的源代码
+程序剖析 | [函数插件] 一键可以剖析其他Python/C++等项目
+读论文 | [函数插件] 一键解读latex论文全文并生成摘要
+arxiv小助手 | [函数插件] 输入url一键翻译摘要+下载论文
+批量注释生成 | [函数插件] 一键批量生成函数注释
+chat分析报告生成 | [函数插件] 运行后自动生成总结汇报
 公式显示 | 可以同时显示公式的tex形式和渲染形式
 图片显示 | 可以在markdown中显示图片
 支持GPT输出的markdown表格 | 可以输出支持GPT的markdown表格
+本地大语言模型接口 | 借助[TGUI](https://github.com/oobabooga/text-generation-webui)接入galactica等本地语言模型
 …… | ……
 
 </div>

config.py

@@ -1,5 +1,5 @@
 # [step 1]>> 例如： API_KEY = "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" （此key无效）
-API_KEY = "sk-此处填API密钥"
+API_KEY = "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"
 
 # [step 2]>> 改为True应用代理，如果直接在海外服务器部署，此处不修改
 USE_PROXY = False
@@ -34,7 +34,7 @@ WEB_PORT = -1
 MAX_RETRY = 2
 
 # OpenAI模型选择是（gpt4现在只对申请成功的人开放）
-LLM_MODEL = "gpt-3.5-turbo"
+LLM_MODEL = "TGUI:galactica-1.3b@localhost:7860" # "gpt-3.5-turbo"
 
 # OpenAI的API_URL
 API_URL = "https://api.openai.com/v1/chat/completions"

crazy_functions/下载arxiv论文翻译摘要.py

@@ -0,0 +1,187 @@
+from predict import predict_no_ui
+from toolbox import CatchException, report_execption, write_results_to_file, predict_no_ui_but_counting_down, get_conf
+import re, requests, unicodedata, os
+
+def download_arxiv_(url_pdf):
+    if 'arxiv.org' not in url_pdf:
+        if ('.' in url_pdf) and ('/' not in url_pdf):
+            new_url = 'https://arxiv.org/abs/'+url_pdf
+            print('下载编号：', url_pdf, '自动定位：', new_url)
+            # download_arxiv_(new_url)
+            return download_arxiv_(new_url)
+        else:
+            print('不能识别的URL！')
+            return None
+    if 'abs' in url_pdf:
+        url_pdf = url_pdf.replace('abs', 'pdf')
+        url_pdf = url_pdf + '.pdf'
+
+    url_abs = url_pdf.replace('.pdf', '').replace('pdf', 'abs')
+    title, other_info = get_name(_url_=url_abs)
+
+    paper_id = title.split()[0]  # '[1712.00559]'
+    if '2' in other_info['year']:
+        title = other_info['year'] + ' ' + title
+
+    known_conf = ['NeurIPS', 'NIPS', 'Nature', 'Science', 'ICLR', 'AAAI']
+    for k in known_conf:
+        if k in other_info['comment']:
+            title = k + ' ' + title
+
+    download_dir = './gpt_log/arxiv/'
+    os.makedirs(download_dir, exist_ok=True)
+
+    title_str = title.replace('?', '？')\
+        .replace(':', '：')\
+        .replace('\"', '“')\
+        .replace('\n', '')\
+        .replace('  ', ' ')\
+        .replace('  ', ' ')
+
+    requests_pdf_url = url_pdf
+    file_path = download_dir+title_str
+    # if os.path.exists(file_path):
+    #     print('返回缓存文件')
+    #     return './gpt_log/arxiv/'+title_str
+
+    print('下载中')
+    proxies, = get_conf('proxies')
+    r = requests.get(requests_pdf_url, proxies=proxies)
+    with open(file_path, 'wb+') as f:
+        f.write(r.content)
+    print('下载完成')
+
+    # print('输出下载命令：','aria2c -o \"%s\" %s'%(title_str,url_pdf))
+    # subprocess.call('aria2c --all-proxy=\"172.18.116.150:11084\" -o \"%s\" %s'%(download_dir+title_str,url_pdf), shell=True)
+
+    x = "%s  %s %s.bib" % (paper_id, other_info['year'], other_info['authors'])
+    x = x.replace('?', '？')\
+        .replace(':', '：')\
+        .replace('\"', '“')\
+        .replace('\n', '')\
+        .replace('  ', ' ')\
+        .replace('  ', ' ')
+    return './gpt_log/arxiv/'+title_str, other_info
+
+
+def get_name(_url_):
+    import os
+    from bs4 import BeautifulSoup
+    print('正在获取文献名！')
+    print(_url_)
+
+    # arxiv_recall = {}
+    # if os.path.exists('./arxiv_recall.pkl'):
+    #     with open('./arxiv_recall.pkl', 'rb') as f:
+    #         arxiv_recall = pickle.load(f)
+
+    # if _url_ in arxiv_recall:
+    #     print('在缓存中')
+    #     return arxiv_recall[_url_]
+
+    proxies, = get_conf('proxies')
+    res = requests.get(_url_, proxies=proxies)
+
+    bs = BeautifulSoup(res.text, 'html.parser')
+    other_details = {}
+
+    # get year
+    try:
+        year = bs.find_all(class_='dateline')[0].text
+        year = re.search(r'(\d{4})', year, re.M | re.I).group(1)
+        other_details['year'] = year
+        abstract = bs.find_all(class_='abstract mathjax')[0].text
+        other_details['abstract'] = abstract
+    except:
+        other_details['year'] = ''
+        print('年份获取失败')
+
+    # get author
+    try:
+        authors = bs.find_all(class_='authors')[0].text
+        authors = authors.split('Authors:')[1]
+        other_details['authors'] = authors
+    except:
+        other_details['authors'] = ''
+        print('authors获取失败')
+
+    # get comment
+    try:
+        comment = bs.find_all(class_='metatable')[0].text
+        real_comment = None
+        for item in comment.replace('\n', ' ').split('   '):
+            if 'Comments' in item:
+                real_comment = item
+        if real_comment is not None:
+            other_details['comment'] = real_comment
+        else:
+            other_details['comment'] = ''
+    except:
+        other_details['comment'] = ''
+        print('年份获取失败')
+
+    title_str = BeautifulSoup(
+        res.text, 'html.parser').find('title').contents[0]
+    print('获取成功：', title_str)
+    # arxiv_recall[_url_] = (title_str+'.pdf', other_details)
+    # with open('./arxiv_recall.pkl', 'wb') as f:
+    #     pickle.dump(arxiv_recall, f)
+
+    return title_str+'.pdf', other_details
+
+
+
+@CatchException
+def 下载arxiv论文并翻译摘要(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+
+    CRAZY_FUNCTION_INFO = "下载arxiv论文并翻译摘要，作者 binary-husky。正在提取摘要并下载PDF文档……"
+    raise RuntimeError()
+    import glob
+    import os
+
+    # 基本信息：功能、贡献者
+    chatbot.append(["函数插件功能？", CRAZY_FUNCTION_INFO])
+    yield chatbot, history, '正常'
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import pdfminer, bs4
+    except:
+        report_execption(chatbot, history, 
+            a = f"解析项目: {txt}", 
+            b = f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade pdfminer beautifulsoup4```。")
+        yield chatbot, history, '正常'
+        return
+
+    # 清空历史，以免输入溢出
+    history = []
+
+    # 提取摘要，下载PDF文档
+    try:
+        pdf_path, info = download_arxiv_(txt)
+    except:
+        report_execption(chatbot, history, 
+            a = f"解析项目: {txt}", 
+            b = f"下载pdf文件未成功")
+        yield chatbot, history, '正常'
+        return
+    
+    # 翻译摘要等
+    i_say =            f"请你阅读以下学术论文相关的材料，提取摘要，翻译为中文。材料如下：{str(info)}"
+    i_say_show_user =  f'请你阅读以下学术论文相关的材料，提取摘要，翻译为中文。论文：{pdf_path}'
+    chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+    yield chatbot, history, '正常'
+    msg = '正常'
+    # ** gpt request **
+    gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temperature, history=[])   # 带超时倒计时
+    chatbot[-1] = (i_say_show_user, gpt_say)
+    history.append(i_say_show_user); history.append(gpt_say)
+    yield chatbot, history, msg
+    # 写入文件
+    import shutil
+    # 重置文件的创建时间
+    shutil.copyfile(pdf_path, pdf_path.replace('.pdf', '.autodownload.pdf')); os.remove(pdf_path)
+    res = write_results_to_file(history)
+    chatbot.append(("完成了吗？", res))
+    yield chatbot, history, msg
+

crazy_functions/解析项目源代码.py

@@ -148,3 +148,20 @@ def 解析一个C项目(txt, top_p, temperature, chatbot, history, systemPromptT
         return
     yield from 解析源代码(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
 
+@CatchException
+def 解析一个Golang项目(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield chatbot, history, '正常'
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.go', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何golang文件: {txt}")
+        yield chatbot, history, '正常'
+        return
+    yield from 解析源代码(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)

functional_crazy.py

@@ -14,6 +14,7 @@ def get_crazy_functionals():
     from crazy_functions.解析项目源代码 import 解析一个Python项目
     from crazy_functions.解析项目源代码 import 解析一个C项目的头文件
     from crazy_functions.解析项目源代码 import 解析一个C项目
+    from crazy_functions.解析项目源代码 import 解析一个Golang项目
     from crazy_functions.高级功能函数模板 import 高阶功能模板函数
     from crazy_functions.代码重写为全英文_多线程 import 全项目切换英文
 
@@ -35,6 +36,11 @@ def get_crazy_functionals():
             "AsButton": False,  # 加入下拉菜单中
             "Function": 解析一个C项目
         },
+        "解析整个Go项目": {
+            "Color": "stop",    # 按钮颜色
+            "AsButton": False,  # 加入下拉菜单中
+            "Function": 解析一个Golang项目
+        },
         "读Tex论文写摘要": {
             "Color": "stop",    # 按钮颜色
             "Function": 读文章写摘要

main.py

@@ -11,8 +11,9 @@ proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION, CHATBOT_HEIGHT =
 PORT = find_free_port() if WEB_PORT <= 0 else WEB_PORT
 if not AUTHENTICATION: AUTHENTICATION = None
 
+title = "ChatGPT 学术优化" if LLM_MODEL.startswith('gpt') else "ChatGPT / LLM 学术优化"
 initial_prompt = "Serve me as a writing and programming assistant."
-title_html = """<h1 align="center">ChatGPT 学术优化</h1>"""
+title_html = f"<h1 align=\"center\">{title}</h1>"
 
 # 问询记录, python 版本建议3.9+（越新越好）
 import logging
@@ -140,5 +141,5 @@ def auto_opentab_delay():
     threading.Thread(target=open, name="open-browser", daemon=True).start()
 
 auto_opentab_delay()
-demo.title = "ChatGPT 学术优化"
+demo.title = title
 demo.queue(concurrency_count=CONCURRENT_COUNT).launch(server_name="0.0.0.0", share=True, server_port=PORT, auth=AUTHENTICATION)

predict.py

@@ -112,8 +112,7 @@ def predict_no_ui_long_connection(inputs, top_p, temperature, history=[], sys_pr
     return result
 
 
-def predict(inputs, top_p, temperature, chatbot=[], history=[], system_prompt='', 
-            stream = True, additional_fn=None):
+def predict(inputs, top_p, temperature, chatbot=[], history=[], system_prompt='', stream = True, additional_fn=None):
     """
         发送至chatGPT，流式获取输出。
         用于基础的对话功能。
@@ -244,3 +243,11 @@ def generate_payload(inputs, top_p, temperature, history, system_prompt, stream)
     return headers,payload
 
 
+if not LLM_MODEL.startswith('gpt'):
+    # 函数重载到另一个文件
+    from request_llm.bridge_tgui import predict_tgui, predict_tgui_no_ui
+    predict = predict_tgui
+    predict_no_ui = predict_tgui_no_ui
+    predict_no_ui_long_connection = predict_tgui_no_ui
+
+    

request_llm/README.md

@@ -0,0 +1,36 @@
+# 如何使用其他大语言模型
+
+## 1. 先运行text-generation
+``` sh
+# 下载模型（ text-generation 这么牛的项目，别忘了给人家star ）
+git clone https://github.com/oobabooga/text-generation-webui.git
+
+# 安装text-generation的额外依赖
+pip install accelerate bitsandbytes flexgen gradio llamacpp markdown numpy peft requests rwkv safetensors sentencepiece tqdm datasets git+https://github.com/huggingface/transformers
+
+# 切换路径
+cd text-generation-webui
+
+# 下载模型
+python download-model.py facebook/galactica-1.3b
+# 其他可选如 facebook/opt-1.3b
+#           facebook/galactica-6.7b
+#           facebook/galactica-120b
+#           facebook/pygmalion-1.3b 等
+# 详情见 https://github.com/oobabooga/text-generation-webui
+
+# 启动text-generation，注意把模型的斜杠改成下划线
+python server.py --cpu --listen --listen-port 7860 --model facebook_galactica-1.3b
+```
+
+## 2. 修改config.py
+``` sh
+# LLM_MODEL格式较复杂   TGUI:[模型]@[ws地址]:[ws端口] ,   端口要和上面给定的端口一致
+LLM_MODEL = "TGUI:galactica-1.3b@localhost:7860"
+```
+
+## 3. 运行！
+``` sh
+cd chatgpt-academic
+python main.py
+```

request_llm/bridge_tgui.py

@@ -0,0 +1,167 @@
+'''
+Contributed by SagsMug. Modified by binary-husky
+https://github.com/oobabooga/text-generation-webui/pull/175
+'''
+
+import asyncio
+import json
+import random
+import string
+import websockets
+import logging
+import time
+import threading
+import importlib
+from toolbox import get_conf
+LLM_MODEL, = get_conf('LLM_MODEL')
+
+# "TGUI:galactica-1.3b@localhost:7860"
+model_name, addr_port = LLM_MODEL.split('@')
+assert ':' in addr_port, "LLM_MODEL 格式不正确！" + LLM_MODEL
+addr, port = addr_port.split(':')
+
+def random_hash():
+    letters = string.ascii_lowercase + string.digits
+    return ''.join(random.choice(letters) for i in range(9))
+
+async def run(context, max_token=512):
+    params = {
+        'max_new_tokens': max_token,
+        'do_sample': True,
+        'temperature': 0.5,
+        'top_p': 0.9,
+        'typical_p': 1,
+        'repetition_penalty': 1.05,
+        'encoder_repetition_penalty': 1.0,
+        'top_k': 0,
+        'min_length': 0,
+        'no_repeat_ngram_size': 0,
+        'num_beams': 1,
+        'penalty_alpha': 0,
+        'length_penalty': 1,
+        'early_stopping': True,
+        'seed': -1,
+    }
+    session = random_hash()
+
+    async with websockets.connect(f"ws://{addr}:{port}/queue/join") as websocket:
+        while content := json.loads(await websocket.recv()):
+            #Python3.10 syntax, replace with if elif on older
+            if content["msg"] ==  "send_hash":
+                await websocket.send(json.dumps({
+                    "session_hash": session,
+                    "fn_index": 12
+                }))
+            elif content["msg"] ==  "estimation":
+                pass
+            elif content["msg"] ==  "send_data":
+                await websocket.send(json.dumps({
+                    "session_hash": session,
+                    "fn_index": 12,
+                    "data": [
+                        context,
+                        params['max_new_tokens'],
+                        params['do_sample'],
+                        params['temperature'],
+                        params['top_p'],
+                        params['typical_p'],
+                        params['repetition_penalty'],
+                        params['encoder_repetition_penalty'],
+                        params['top_k'],
+                        params['min_length'],
+                        params['no_repeat_ngram_size'],
+                        params['num_beams'],
+                        params['penalty_alpha'],
+                        params['length_penalty'],
+                        params['early_stopping'],
+                        params['seed'],
+                    ]
+                }))
+            elif content["msg"] ==  "process_starts":
+                pass
+            elif content["msg"] in ["process_generating", "process_completed"]:
+                yield content["output"]["data"][0]
+                # You can search for your desired end indicator and 
+                #  stop generation by closing the websocket here
+                if (content["msg"] == "process_completed"):
+                    break
+
+
+
+
+
+def predict_tgui(inputs, top_p, temperature, chatbot=[], history=[], system_prompt='', stream = True, additional_fn=None):
+    """
+        发送至chatGPT，流式获取输出。
+        用于基础的对话功能。
+        inputs 是本次问询的输入
+        top_p, temperature是chatGPT的内部调优参数
+        history 是之前的对话列表（注意无论是inputs还是history，内容太长了都会触发token数量溢出的错误）
+        chatbot 为WebUI中显示的对话列表，修改它，然后yeild出去，可以直接修改对话界面内容
+        additional_fn代表点击的哪个按钮，按钮见functional.py
+    """
+    if additional_fn is not None:
+        import functional
+        importlib.reload(functional)    # 热更新prompt
+        functional = functional.get_functionals()
+        if "PreProcess" in functional[additional_fn]: inputs = functional[additional_fn]["PreProcess"](inputs)  # 获取预处理函数（如果有的话）
+        inputs = functional[additional_fn]["Prefix"] + inputs + functional[additional_fn]["Suffix"]
+
+    raw_input = "What I would like to say is the following: " + inputs
+    logging.info(f'[raw_input] {raw_input}')
+    history.extend([inputs, ""])
+    chatbot.append([inputs, ""])
+    yield chatbot, history, "等待响应"
+
+    prompt = inputs
+    tgui_say = ""
+
+    mutable = ["", time.time()]
+    def run_coorotine(mutable):
+        async def get_result(mutable):
+            async for response in run(prompt):
+                print(response[len(mutable[0]):])
+                mutable[0] = response
+                if (time.time() - mutable[1]) > 3: 
+                    print('exit when no listener')
+                    break
+        asyncio.run(get_result(mutable))
+
+    thread_listen = threading.Thread(target=run_coorotine, args=(mutable,), daemon=True)
+    thread_listen.start()
+
+    while thread_listen.is_alive():
+        time.sleep(1)
+        mutable[1] = time.time()
+        # Print intermediate steps
+        if tgui_say != mutable[0]:
+            tgui_say = mutable[0]
+            history[-1] = tgui_say
+            chatbot[-1] = (history[-2], history[-1])
+            yield chatbot, history, "status_text"
+
+    logging.info(f'[response] {tgui_say}')
+
+
+
+def predict_tgui_no_ui(inputs, top_p, temperature, history=[], sys_prompt=""):
+    raw_input = "What I would like to say is the following: " + inputs
+    prompt = inputs
+    tgui_say = ""
+    mutable = ["", time.time()]
+    def run_coorotine(mutable):
+        async def get_result(mutable):
+            async for response in run(prompt, max_token=20):
+                print(response[len(mutable[0]):])
+                mutable[0] = response
+                if (time.time() - mutable[1]) > 3: 
+                    print('exit when no listener')
+                    break
+        asyncio.run(get_result(mutable))
+    thread_listen = threading.Thread(target=run_coorotine, args=(mutable,))
+    thread_listen.start()
+    while thread_listen.is_alive():
+        time.sleep(1)
+        mutable[1] = time.time()
+    tgui_say = mutable[0]
+    return tgui_say