Version 3.56 - Merge branch 'frontier'

This commit is contained in:
binary-husky
2023-10-22 23:24:41 +08:00
21 changed files with 727 additions and 65 deletions

View File

@@ -0,0 +1,23 @@
from toolbox import CatchException, update_ui, gen_time_str, trimmed_format_exc, ProxyNetworkActivate
from toolbox import report_execption, get_log_folder, update_ui_lastest_msg, Singleton
from crazy_functions.agent_fns.pipe import PluginMultiprocessManager, PipeCom
from crazy_functions.agent_fns.autogen_general import AutoGenGeneral
import time
class AutoGenMath(AutoGenGeneral):
def define_agents(self):
from autogen import AssistantAgent, UserProxyAgent
return [
{
"name": "assistant", # name of the agent.
"cls": AssistantAgent, # class of the agent.
},
{
"name": "user_proxy", # name of the agent.
"cls": UserProxyAgent, # class of the agent.
"human_input_mode": "ALWAYS", # always ask for human input.
"llm_config": False, # disables llm-based auto reply.
},
]

View File

@@ -0,0 +1,75 @@
from toolbox import CatchException, update_ui, gen_time_str, trimmed_format_exc, ProxyNetworkActivate
from toolbox import report_execption, get_log_folder, update_ui_lastest_msg, Singleton
from crazy_functions.agent_fns.pipe import PluginMultiprocessManager, PipeCom
import time
class AutoGenGeneral(PluginMultiprocessManager):
def gpt_academic_print_override(self, user_proxy, message, sender):
# ⭐⭐ 子进程执行
self.child_conn.send(PipeCom("show", sender.name + '\n\n---\n\n' + message['content']))
def gpt_academic_get_human_input(self, user_proxy, message):
# ⭐⭐ 子进程执行
patience = 300
begin_waiting_time = time.time()
self.child_conn.send(PipeCom("interact", message))
while True:
time.sleep(0.5)
if self.child_conn.poll():
wait_success = True
break
if time.time() - begin_waiting_time > patience:
self.child_conn.send(PipeCom("done", ""))
wait_success = False
break
if wait_success:
return self.child_conn.recv().content
else:
raise TimeoutError("等待用户输入超时")
def define_agents(self):
raise NotImplementedError
def do_audogen(self, input):
# ⭐⭐ 子进程执行
input = input.content
with ProxyNetworkActivate("AutoGen"):
from autogen import AssistantAgent, UserProxyAgent
config_list = [{
'model': self.llm_kwargs['llm_model'],
'api_key': self.llm_kwargs['api_key'],
},]
code_execution_config={"work_dir": self.autogen_work_dir, "use_docker":True}
agents = self.define_agents()
user_proxy = None
assistant = None
for agent_kwargs in agents:
agent_cls = agent_kwargs.pop('cls')
kwargs = {
'llm_config':{
"config_list": config_list,
},
'code_execution_config':code_execution_config
}
kwargs.update(agent_kwargs)
agent_handle = agent_cls(**kwargs)
agent_handle._print_received_message = lambda a,b: self.gpt_academic_print_override(agent_kwargs, a, b)
if agent_kwargs['name'] == 'user_proxy':
agent_handle.get_human_input = lambda a: self.gpt_academic_get_human_input(user_proxy, a)
user_proxy = agent_handle
if agent_kwargs['name'] == 'assistant': assistant = agent_handle
try:
if user_proxy is None or assistant is None: raise Exception("用户代理或助理代理未定义")
user_proxy.initiate_chat(assistant, message=input)
except Exception as e:
tb_str = '```\n' + trimmed_format_exc() + '```'
self.child_conn.send(PipeCom("done", "AutoGen 执行失败: \n\n" + tb_str))
def subprocess_worker(self, child_conn):
# ⭐⭐ 子进程执行
self.child_conn = child_conn
while True:
msg = self.child_conn.recv() # PipeCom
self.do_audogen(msg)

View File

@@ -0,0 +1,19 @@
from crazy_functions.agent_fns.pipe import PluginMultiprocessManager, PipeCom
class EchoDemo(PluginMultiprocessManager):
def subprocess_worker(self, child_conn):
# ⭐⭐ 子进程
self.child_conn = child_conn
while True:
msg = self.child_conn.recv() # PipeCom
if msg.cmd == "user_input":
# wait futher user input
self.child_conn.send(PipeCom("show", msg.content))
wait_success = self.subprocess_worker_wait_user_feedback(wait_msg="我准备好处理下一个问题了.")
if not wait_success:
# wait timeout, terminate this subprocess_worker
break
elif msg.cmd == "terminate":
self.child_conn.send(PipeCom("done", ""))
break
print('[debug] subprocess_worker terminated')

View File

@@ -0,0 +1,16 @@
from toolbox import Singleton
@Singleton
class GradioMultiuserManagerForPersistentClasses():
def __init__(self):
self.mapping = {}
def already_alive(self, key):
return (key in self.mapping) and (self.mapping[key].is_alive())
def set(self, key, x):
self.mapping[key] = x
return self.mapping[key]
def get(self, key):
return self.mapping[key]

View File

@@ -0,0 +1,150 @@
from toolbox import get_log_folder, update_ui, gen_time_str, trimmed_format_exc, promote_file_to_downloadzone
import time, os
class PipeCom():
def __init__(self, cmd, content) -> None:
self.cmd = cmd
self.content = content
class PluginMultiprocessManager():
def __init__(self, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
# ⭐ 主进程
self.autogen_work_dir = os.path.join(get_log_folder('autogen'), gen_time_str())
self.previous_work_dir_files = {}
self.llm_kwargs = llm_kwargs
self.plugin_kwargs = plugin_kwargs
self.chatbot = chatbot
self.history = history
self.system_prompt = system_prompt
self.web_port = web_port
self.alive = True
def is_alive(self):
return self.alive
def launch_subprocess_with_pipe(self):
# ⭐ 主进程
from multiprocessing import Process, Pipe
parent_conn, child_conn = Pipe()
self.p = Process(target=self.subprocess_worker, args=(child_conn,))
self.p.daemon = True
self.p.start()
return parent_conn
def terminate(self):
self.p.terminate()
self.alive = False
print('[debug] instance terminated')
def subprocess_worker(self, child_conn):
# ⭐⭐ 子进程
raise NotImplementedError
def send_command(self, cmd):
# ⭐ 主进程
self.parent_conn.send(PipeCom("user_input", cmd))
def immediate_showoff_when_possible(self, fp):
# ⭐ 主进程
# 获取fp的拓展名
file_type = fp.split('.')[-1]
# 如果是文本文件, 则直接显示文本内容
if file_type in ['png', 'jpg']:
image_path = os.path.abspath(fp)
self.chatbot.append(['检测到新生图像:', f'本地文件预览: <br/><div align="center"><img src="file={image_path}"></div>'])
yield from update_ui(chatbot=self.chatbot, history=self.history)
def overwatch_workdir_file_change(self):
# ⭐ 主进程 Docker 外挂文件夹监控
path_to_overwatch = self.autogen_work_dir
change_list = []
# 扫描路径下的所有文件, 并与self.previous_work_dir_files中所记录的文件进行对比
# 如果有新文件出现或者文件的修改时间发生变化则更新self.previous_work_dir_files中
# 把新文件和发生变化的文件的路径记录到 change_list 中
for root, dirs, files in os.walk(path_to_overwatch):
for file in files:
file_path = os.path.join(root, file)
if file_path not in self.previous_work_dir_files.keys():
last_modified_time = os.stat(file_path).st_mtime
self.previous_work_dir_files.update({file_path:last_modified_time})
change_list.append(file_path)
else:
last_modified_time = os.stat(file_path).st_mtime
if last_modified_time != self.previous_work_dir_files[file_path]:
self.previous_work_dir_files[file_path] = last_modified_time
change_list.append(file_path)
if len(change_list) > 0:
file_links = ''
for f in change_list:
res = promote_file_to_downloadzone(f)
file_links += f'<br/><a href="file={res}" target="_blank">{res}</a>'
yield from self.immediate_showoff_when_possible(file_path)
self.chatbot.append(['检测到新生文档.', f'文档清单如下: {file_links}'])
yield from update_ui(chatbot=self.chatbot, history=self.history)
return change_list
def main_process_ui_control(self, txt, create_or_resume) -> str:
# ⭐ 主进程
if create_or_resume == 'create':
self.cnt = 1
self.parent_conn = self.launch_subprocess_with_pipe() # ⭐⭐⭐
self.send_command(txt)
if txt == 'exit':
self.chatbot.append([f"结束", "结束信号已明确终止AutoGen程序。"])
yield from update_ui(chatbot=self.chatbot, history=self.history)
self.terminate()
return "terminate"
while True:
time.sleep(0.5)
if self.parent_conn.poll():
if '[GPT-Academic] 等待中' in self.chatbot[-1][-1]:
self.chatbot.pop(-1) # remove the last line
msg = self.parent_conn.recv() # PipeCom
if msg.cmd == "done":
self.chatbot.append([f"结束", msg.content]); self.cnt += 1
yield from update_ui(chatbot=self.chatbot, history=self.history)
self.terminate(); break
if msg.cmd == "show":
yield from self.overwatch_workdir_file_change()
self.chatbot.append([f"运行阶段-{self.cnt}", msg.content]); self.cnt += 1
yield from update_ui(chatbot=self.chatbot, history=self.history)
if msg.cmd == "interact":
yield from self.overwatch_workdir_file_change()
self.chatbot.append([f"程序抵达用户反馈节点.", msg.content +
"\n\n等待您的进一步指令." +
"\n\n(1) 一般情况下您不需要说什么, 清空输入区, 然后直接点击“提交”以继续. " +
"\n\n(2) 如果您需要补充些什么, 输入要反馈的内容, 直接点击“提交”以继续. " +
"\n\n(3) 如果您想终止程序, 输入exit, 直接点击“提交”以终止AutoGen并解锁. "
])
yield from update_ui(chatbot=self.chatbot, history=self.history)
# do not terminate here, leave the subprocess_worker instance alive
return "wait_feedback"
else:
if '[GPT-Academic] 等待中' not in self.chatbot[-1][-1]:
self.chatbot.append(["[GPT-Academic] 等待AutoGen执行结果 ...", "[GPT-Academic] 等待中"])
self.chatbot[-1] = [self.chatbot[-1][0], self.chatbot[-1][1].replace("[GPT-Academic] 等待中", "[GPT-Academic] 等待中.")]
yield from update_ui(chatbot=self.chatbot, history=self.history)
self.terminate()
return "terminate"
def subprocess_worker_wait_user_feedback(self, wait_msg="wait user feedback"):
# ⭐⭐ 子进程
patience = 5 * 60
begin_waiting_time = time.time()
self.child_conn.send(PipeCom("interact", wait_msg))
while True:
time.sleep(0.5)
if self.child_conn.poll():
wait_success = True
break
if time.time() - begin_waiting_time > patience:
self.child_conn.send(PipeCom("done", ""))
wait_success = False
break
return wait_success

View File

@@ -721,8 +721,10 @@ class nougat_interface():
def nougat_with_timeout(self, command, cwd, timeout=3600):
import subprocess
from toolbox import ProxyNetworkActivate
logging.info(f'正在执行命令 {command}')
process = subprocess.Popen(command, shell=True, cwd=cwd)
with ProxyNetworkActivate("Nougat_Download"):
process = subprocess.Popen(command, shell=True, cwd=cwd, env=os.environ)
try:
stdout, stderr = process.communicate(timeout=timeout)
except subprocess.TimeoutExpired:
@@ -767,54 +769,6 @@ def try_install_deps(deps, reload_m=[]):
importlib.reload(__import__(m))
HTML_CSS = """
.row {
display: flex;
flex-wrap: wrap;
}
.column {
flex: 1;
padding: 10px;
}
.table-header {
font-weight: bold;
border-bottom: 1px solid black;
}
.table-row {
border-bottom: 1px solid lightgray;
}
.table-cell {
padding: 5px;
}
"""
TABLE_CSS = """
<div class="row table-row">
<div class="column table-cell">REPLACE_A</div>
<div class="column table-cell">REPLACE_B</div>
</div>
"""
class construct_html():
def __init__(self) -> None:
self.css = HTML_CSS
self.html_string = f'<!DOCTYPE html><head><meta charset="utf-8"><title>翻译结果</title><style>{self.css}</style></head>'
def add_row(self, a, b):
tmp = TABLE_CSS
from toolbox import markdown_convertion
tmp = tmp.replace('REPLACE_A', markdown_convertion(a))
tmp = tmp.replace('REPLACE_B', markdown_convertion(b))
self.html_string += tmp
def save_file(self, file_name):
with open(os.path.join(get_log_folder(), file_name), 'w', encoding='utf8') as f:
f.write(self.html_string.encode('utf-8', 'ignore').decode())
return os.path.join(get_log_folder(), file_name)
def get_plugin_arg(plugin_kwargs, key, default):
# 如果参数是空的
if (key in plugin_kwargs) and (plugin_kwargs[key] == ""): plugin_kwargs.pop(key)

View File

@@ -423,7 +423,7 @@ def write_html(sp_file_contents, sp_file_result, chatbot, project_folder):
# write html
try:
import shutil
from ..crazy_utils import construct_html
from crazy_functions.pdf_fns.report_gen_html import construct_html
from toolbox import gen_time_str
ch = construct_html()
orig = ""

View File

@@ -308,7 +308,10 @@ def merge_tex_files_(project_foler, main_file, mode):
fp = os.path.join(project_foler, f)
fp_ = find_tex_file_ignore_case(fp)
if fp_:
with open(fp_, 'r', encoding='utf-8', errors='replace') as fx: c = fx.read()
try:
with open(fp_, 'r', encoding='utf-8', errors='replace') as fx: c = fx.read()
except:
c = f"\n\nWarning from GPT-Academic: LaTex source file is missing!\n\n"
else:
raise RuntimeError(f'找不到{fp}Tex源文件缺失')
c = merge_tex_files_(project_foler, c, mode)
@@ -366,6 +369,14 @@ def insert_abstract(tex_content):
# insert "abs_str" on the next line
modified_tex = tex_content[:end_line_index+1] + '\n\n' + insert_missing_abs_str + '\n\n' + tex_content[end_line_index+1:]
return modified_tex
elif r"\begin{document}" in tex_content:
# find the position of "\maketitle"
find_index = tex_content.index(r"\begin{document}")
# find the nearest ending line
end_line_index = tex_content.find("\n", find_index)
# insert "abs_str" on the next line
modified_tex = tex_content[:end_line_index+1] + '\n\n' + insert_missing_abs_str + '\n\n' + tex_content[end_line_index+1:]
return modified_tex
else:
return tex_content

View File

@@ -73,7 +73,7 @@ def produce_report_markdown(gpt_response_collection, meta, paper_meta_info, chat
return res_path
def translate_pdf(article_dict, llm_kwargs, chatbot, fp, generated_conclusion_files, TOKEN_LIMIT_PER_FRAGMENT, DST_LANG):
from crazy_functions.crazy_utils import construct_html
from crazy_functions.pdf_fns.report_gen_html import construct_html
from crazy_functions.crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
from crazy_functions.crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
from crazy_functions.crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency

View File

@@ -0,0 +1,58 @@
from toolbox import update_ui, get_conf, trimmed_format_exc, get_log_folder
import os
class construct_html():
def __init__(self) -> None:
self.html_string = ""
def add_row(self, a, b):
from toolbox import markdown_convertion
template = """
{
primary_col: {
header: String.raw`__PRIMARY_HEADER__`,
msg: String.raw`__PRIMARY_MSG__`,
},
secondary_rol: {
header: String.raw`__SECONDARY_HEADER__`,
msg: String.raw`__SECONDARY_MSG__`,
}
},
"""
def std(str):
str = str.replace(r'`',r'&#96;')
if str.endswith("\\"): str += ' '
if str.endswith("}"): str += ' '
if str.endswith("$"): str += ' '
return str
template_ = template
a_lines = a.split('\n')
b_lines = b.split('\n')
if len(a_lines) == 1 or len(a_lines[0]) > 50:
template_ = template_.replace("__PRIMARY_HEADER__", std(a[:20]))
template_ = template_.replace("__PRIMARY_MSG__", std(markdown_convertion(a)))
else:
template_ = template_.replace("__PRIMARY_HEADER__", std(a_lines[0]))
template_ = template_.replace("__PRIMARY_MSG__", std(markdown_convertion('\n'.join(a_lines[1:]))))
if len(b_lines) == 1 or len(b_lines[0]) > 50:
template_ = template_.replace("__SECONDARY_HEADER__", std(b[:20]))
template_ = template_.replace("__SECONDARY_MSG__", std(markdown_convertion(b)))
else:
template_ = template_.replace("__SECONDARY_HEADER__", std(b_lines[0]))
template_ = template_.replace("__SECONDARY_MSG__", std(markdown_convertion('\n'.join(b_lines[1:]))))
self.html_string += template_
def save_file(self, file_name):
from toolbox import get_log_folder
with open('crazy_functions/pdf_fns/report_template.html', 'r', encoding='utf8') as f:
html_template = f.read()
html_template = html_template.replace("__TF_ARR__", self.html_string)
with open(os.path.join(get_log_folder(), file_name), 'w', encoding='utf8') as f:
f.write(html_template.encode('utf-8', 'ignore').decode())
return os.path.join(get_log_folder(), file_name)

File diff suppressed because one or more lines are too long

View File

@@ -0,0 +1,96 @@
# 本源代码中, ⭐ = 关键步骤
"""
测试:
- 裁剪图像,保留下半部分
- 交换图像的蓝色通道和红色通道
- 将图像转为灰度图像
- 将csv文件转excel表格
Testing:
- Crop the image, keeping the bottom half.
- Swap the blue channel and red channel of the image.
- Convert the image to grayscale.
- Convert the CSV file to an Excel spreadsheet.
"""
from toolbox import CatchException, update_ui, gen_time_str, trimmed_format_exc, ProxyNetworkActivate
from toolbox import get_conf, select_api_key, update_ui_lastest_msg, Singleton
from crazy_functions.crazy_utils import request_gpt_model_in_new_thread_with_ui_alive, get_plugin_arg
from crazy_functions.crazy_utils import input_clipping, try_install_deps
from crazy_functions.agent_fns.persistent import GradioMultiuserManagerForPersistentClasses
from crazy_functions.agent_fns.auto_agent import AutoGenMath
import time
@CatchException
def 多智能体终端(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
"""
txt 输入栏用户输入的文本,例如需要翻译的一段话,再例如一个包含了待处理文件的路径
llm_kwargs gpt模型参数如温度和top_p等一般原样传递下去就行
plugin_kwargs 插件模型的参数
chatbot 聊天显示框的句柄,用于显示给用户
history 聊天历史,前情提要
system_prompt 给gpt的静默提醒
web_port 当前软件运行的端口号
"""
# 检查当前的模型是否符合要求
supported_llms = ['gpt-3.5-turbo-16k', 'gpt-4', 'gpt-4-32k']
llm_kwargs['api_key'] = select_api_key(llm_kwargs['api_key'], llm_kwargs['llm_model'])
if llm_kwargs['llm_model'] not in supported_llms:
chatbot.append([f"处理任务: {txt}", f"当前插件只支持{str(supported_llms)}, 当前模型{llm_kwargs['llm_model']}."])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
return
# 检查当前的模型是否符合要求
API_URL_REDIRECT, = get_conf('API_URL_REDIRECT')
if len(API_URL_REDIRECT) > 0:
chatbot.append([f"处理任务: {txt}", f"暂不支持中转."])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
return
# 尝试导入依赖,如果缺少依赖,则给出安装建议
try:
import autogen, docker
except:
chatbot.append([ f"处理任务: {txt}",
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 pyautogen docker```。"])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
return
# 尝试导入依赖,如果缺少依赖,则给出安装建议
try:
import autogen
import glob, os, time, subprocess
subprocess.Popen(['docker', '--version'])
except:
chatbot.append([f"处理任务: {txt}", f"缺少docker运行环境"])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
return
# 解锁插件
chatbot.get_cookies()['lock_plugin'] = None
persistent_class_multi_user_manager = GradioMultiuserManagerForPersistentClasses()
user_uuid = chatbot.get_cookies().get('uuid')
persistent_key = f"{user_uuid}->多智能体终端"
if persistent_class_multi_user_manager.already_alive(persistent_key):
# 当已经存在一个正在运行的多智能体终端时,直接将用户输入传递给它,而不是再次启动一个新的多智能体终端
print('[debug] feed new user input')
executor = persistent_class_multi_user_manager.get(persistent_key)
exit_reason = yield from executor.main_process_ui_control(txt, create_or_resume="resume")
else:
# 运行多智能体终端 (首次)
print('[debug] create new executor instance')
history = []
chatbot.append(["正在启动: 多智能体终端", "插件动态生成, 执行开始, 作者 Microsoft & Binary-Husky."])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
executor = AutoGenMath(llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port)
persistent_class_multi_user_manager.set(persistent_key, executor)
exit_reason = yield from executor.main_process_ui_control(txt, create_or_resume="create")
if exit_reason == "wait_feedback":
# 当用户点击了“等待反馈”按钮时将executor存储到cookie中等待用户的再次调用
executor.chatbot.get_cookies()['lock_plugin'] = 'crazy_functions.多智能体->多智能体终端'
else:
executor.chatbot.get_cookies()['lock_plugin'] = None
yield from update_ui(chatbot=executor.chatbot, history=executor.history) # 更新状态

View File

@@ -97,7 +97,8 @@ def 解析PDF_基于NOUGAT(file_manifest, project_folder, llm_kwargs, plugin_kwa
generated_conclusion_files = []
generated_html_files = []
DST_LANG = "中文"
from crazy_functions.crazy_utils import nougat_interface, construct_html
from crazy_functions.crazy_utils import nougat_interface
from crazy_functions.pdf_fns.report_gen_html import construct_html
nougat_handle = nougat_interface()
for index, fp in enumerate(file_manifest):
chatbot.append(["当前进度:", f"正在解析论文请稍候。第一次运行时需要花费较长时间下载NOUGAT参数"]); yield from update_ui(chatbot=chatbot, history=history) # 刷新界面

View File

@@ -63,7 +63,7 @@ def 解析PDF_基于GROBID(file_manifest, project_folder, llm_kwargs, plugin_kwa
generated_conclusion_files = []
generated_html_files = []
DST_LANG = "中文"
from crazy_functions.crazy_utils import construct_html
from crazy_functions.pdf_fns.report_gen_html import construct_html
for index, fp in enumerate(file_manifest):
chatbot.append(["当前进度:", f"正在连接GROBID服务请稍候: {grobid_url}\n如果等待时间过长请修改config中的GROBID_URL可修改成本地GROBID服务。"]); yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
article_dict = parse_pdf(fp, grobid_url)
@@ -86,7 +86,7 @@ def 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot,
TOKEN_LIMIT_PER_FRAGMENT = 1024
generated_conclusion_files = []
generated_html_files = []
from crazy_functions.crazy_utils import construct_html
from crazy_functions.pdf_fns.report_gen_html import construct_html
for index, fp in enumerate(file_manifest):
# 读取PDF文件
file_content, page_one = read_and_clean_pdf_text(fp)