int/gpt_academic
1
0
派生 0
镜像自地址 https://github.com/binary-husky/gpt_academic.git 已同步 2025-12-06 06:26:47 +00:00
代码 工单 软件包 项目 版本发布 百科 动态

new

浏览代码
这个提交包含在：
binary-husky
2023-12-26 23:59:53 +08:00
父节点 8dd4d48474
当前提交 d245958dfa
共有 17 个文件被更改，包括 1270 次插入 和 0 次删除
下载 Patch 文件 下载 Diff 文件 展开所有文件 折叠所有文件

42
crazy_functions/game_fns/game_ascii_art.py 普通文件
查看文件

@@ -0,0 +1,42 @@
from toolbox import CatchException, update_ui, update_ui_lastest_msg
from crazy_functions.multi_stage.multi_stage_utils import GptAcademicGameBaseState
from crazy_functions.crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
from request_llms.bridge_all import predict_no_ui_long_connection
from crazy_functions.game_fns.game_utils import get_code_block, is_same_thing
import random
class MiniGame_ASCII_Art(GptAcademicGameBaseState):
def step(self, prompt, chatbot, history):
if self.step_cnt == 0:
chatbot.append(["我画你猜（动物）", "请稍等..."])
else:
if prompt.strip() == 'exit':
self.delete_game = True
yield from update_ui_lastest_msg(lastmsg=f"谜底是{self.obj}，游戏结束。", chatbot=chatbot, history=history, delay=0.)
return
chatbot.append([prompt, ""])
yield from update_ui(chatbot=chatbot, history=history)
if self.step_cnt == 0:
self.lock_plugin(chatbot)
self.cur_task = 'draw'
if self.cur_task == 'draw':
avail_obj = ["狗","猫","鸟","鱼","老鼠","蛇"]
self.obj = random.choice(avail_obj)
inputs = "I want to play a game called Guess the ASCII art. You can draw the ASCII art and I will try to guess it. " + \
f"This time you draw a {self.obj}. Note that you must not indicate what you have draw in the text, and you should only produce the ASCII art wrapped by ```. "
raw_res = predict_no_ui_long_connection(inputs=inputs, llm_kwargs=self.llm_kwargs, history=[], sys_prompt="")
self.cur_task = 'identify user guess'
res = get_code_block(raw_res)
history += ['', f'the answer is {self.obj}', inputs, res]
yield from update_ui_lastest_msg(lastmsg=res, chatbot=chatbot, history=history, delay=0.)
elif self.cur_task == 'identify user guess':
if is_same_thing(self.obj, prompt, self.llm_kwargs):
self.delete_game = True
yield from update_ui_lastest_msg(lastmsg="你猜对了！", chatbot=chatbot, history=history, delay=0.)
else:
self.cur_task = 'identify user guess'
yield from update_ui_lastest_msg(lastmsg="猜错了，再试试，输入“exit”获取答案。", chatbot=chatbot, history=history, delay=0.)

212
crazy_functions/game_fns/game_interactive_story.py 普通文件
查看文件

@@ -0,0 +1,212 @@
prompts_hs = """ 请以“{headstart}”为开头，编写一个小说的第一幕。
- 尽量短，不要包含太多情节，因为你接下来将会与用户互动续写下面的情节，要留出足够的互动空间。
- 出现人物时，给出人物的名字。
- 积极地运用环境描写、人物描写等手法，让读者能够感受到你的故事世界。
- 积极地运用修辞手法，比如比喻、拟人、排比、对偶、夸张等等。
- 字数要求：第一幕的字数少于300字，且少于2个段落。
"""
prompts_interact = """ 小说的前文回顾：
「
{previously_on_story}
」
你是一个作家，根据以上的情节，给出4种不同的后续剧情发展方向，每个发展方向都精明扼要地用一句话说明。稍后，我将在这4个选择中，挑选一种剧情发展。
输出格式例如：
1. 后续剧情发展1
2. 后续剧情发展2
3. 后续剧情发展3
4. 后续剧情发展4
"""
prompts_resume = """小说的前文回顾：
「
{previously_on_story}
」
你是一个作家，我们正在互相讨论，确定后续剧情的发展。
在以下的剧情发展中，
「
{choice}
」
我认为更合理的是：{user_choice}。
请在前文的基础上（不要重复前文），围绕我选定的剧情情节，编写小说的下一幕。
- 禁止杜撰不符合我选择的剧情。
- 尽量短，不要包含太多情节，因为你接下来将会与用户互动续写下面的情节，要留出足够的互动空间。
- 不要重复前文。
- 出现人物时，给出人物的名字。
- 积极地运用环境描写、人物描写等手法，让读者能够感受到你的故事世界。
- 积极地运用修辞手法，比如比喻、拟人、排比、对偶、夸张等等。
- 小说的下一幕字数少于300字，且少于2个段落。
"""
prompts_terminate = """小说的前文回顾：
「
{previously_on_story}
」
你是一个作家，我们正在互相讨论，确定后续剧情的发展。
现在，故事该结束了，我认为最合理的故事结局是：{user_choice}。
请在前文的基础上（不要重复前文），编写小说的最后一幕。
- 不要重复前文。
- 出现人物时，给出人物的名字。
- 积极地运用环境描写、人物描写等手法，让读者能够感受到你的故事世界。
- 积极地运用修辞手法，比如比喻、拟人、排比、对偶、夸张等等。
- 字数要求：最后一幕的字数少于1000字。
"""
from toolbox import CatchException, update_ui, update_ui_lastest_msg
from crazy_functions.multi_stage.multi_stage_utils import GptAcademicGameBaseState
from crazy_functions.crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
from request_llms.bridge_all import predict_no_ui_long_connection
from crazy_functions.game_fns.game_utils import get_code_block, is_same_thing
import random
class MiniGame_ResumeStory(GptAcademicGameBaseState):
story_headstart = [
'先行者知道，他现在是全宇宙中唯一的一个人了。',
'深夜，一个年轻人穿过天安门广场向纪念堂走去。在二十二世纪编年史中，计算机把他的代号定为M102。',
'他知道，这最后一课要提前讲了。又一阵剧痛从肝部袭来，几乎使他晕厥过去。',
'在距地球五万光年的远方，在银河系的中心，一场延续了两万年的星际战争已接近尾声。那里的太空中渐渐隐现出一个方形区域，仿佛灿烂的群星的背景被剪出一个方口。',
'伊依一行三人乘坐一艘游艇在南太平洋上做吟诗航行，他们的目的地是南极，如果几天后能顺利到达那里，他们将钻出地壳去看诗云。',
'很多人生来就会莫名其妙地迷上一样东西，仿佛他的出生就是要和这东西约会似的，正是这样，圆圆迷上了肥皂泡。'
]
def begin_game_step_0(self, prompt, chatbot, history):
# init game at step 0
self.headstart = random.choice(self.story_headstart)
self.story = []
chatbot.append(["互动写故事", f"这次的故事开头是：{self.headstart}"])
self.sys_prompt_ = '你是一个想象力丰富的杰出作家。正在与你的朋友互动，一起写故事，因此你每次写的故事段落应少于300字（结局除外）。'
def generate_story_image(self, story_paragraph):
try:
from crazy_functions.图片生成 import gen_image
prompt_ = predict_no_ui_long_connection(inputs=story_paragraph, llm_kwargs=self.llm_kwargs, history=[], sys_prompt='你需要根据用户给出的小说段落，进行简短的环境描写。要求：80字以内。')
image_url, image_path = gen_image(self.llm_kwargs, prompt_, '512x512', model="dall-e-2", quality='standard', style='natural')
return f'<br/><div align="center"><img src="file={image_path}"></div>'
except:
return ''
def step(self, prompt, chatbot, history):
"""
首先，处理游戏初始化等特殊情况
"""
if self.step_cnt == 0:
self.begin_game_step_0(prompt, chatbot, history)
self.lock_plugin(chatbot)
self.cur_task = 'head_start'
else:
if prompt.strip() == 'exit' or prompt.strip() == '结束剧情':
# should we terminate game here?
self.delete_game = True
yield from update_ui_lastest_msg(lastmsg=f"游戏结束。", chatbot=chatbot, history=history, delay=0.)
return
if '剧情收尾' in prompt:
self.cur_task = 'story_terminate'
# # well, game resumes
# chatbot.append([prompt, ""])
# update ui, don't keep the user waiting
yield from update_ui(chatbot=chatbot, history=history)
"""
处理游戏的主体逻辑
"""
if self.cur_task == 'head_start':
"""
这是游戏的第一步
"""
inputs_ = prompts_hs.format(headstart=self.headstart)
history_ = []
story_paragraph = yield from request_gpt_model_in_new_thread_with_ui_alive(
inputs_, '故事开头', self.llm_kwargs,
chatbot, history_, self.sys_prompt_
)
self.story.append(story_paragraph)
# # 配图
yield from update_ui_lastest_msg(lastmsg=story_paragraph + '<br/>正在生成插图中 ...', chatbot=chatbot, history=history, delay=0.)
yield from update_ui_lastest_msg(lastmsg=story_paragraph + '<br/>'+ self.generate_story_image(story_paragraph), chatbot=chatbot, history=history, delay=0.)
# # 构建后续剧情引导
previously_on_story = ""
for s in self.story:
previously_on_story += s + '\n'
inputs_ = prompts_interact.format(previously_on_story=previously_on_story)
history_ = []
self.next_choices = yield from request_gpt_model_in_new_thread_with_ui_alive(
inputs_, '请在以下几种故事走向中，选择一种（当然，您也可以选择给出其他故事走向）：', self.llm_kwargs,
chatbot,
history_,
self.sys_prompt_
)
self.cur_task = 'user_choice'
elif self.cur_task == 'user_choice':
"""
根据用户的提示，确定故事的下一步
"""
if '请在以下几种故事走向中，选择一种' in chatbot[-1][0]: chatbot.pop(-1)
previously_on_story = ""
for s in self.story:
previously_on_story += s + '\n'
inputs_ = prompts_resume.format(previously_on_story=previously_on_story, choice=self.next_choices, user_choice=prompt)
history_ = []
story_paragraph = yield from request_gpt_model_in_new_thread_with_ui_alive(
inputs_, f'下一段故事（您的选择是：{prompt}）。', self.llm_kwargs,
chatbot, history_, self.sys_prompt_
)
self.story.append(story_paragraph)
# # 配图
yield from update_ui_lastest_msg(lastmsg=story_paragraph + '<br/>正在生成插图中 ...', chatbot=chatbot, history=history, delay=0.)
yield from update_ui_lastest_msg(lastmsg=story_paragraph + '<br/>'+ self.generate_story_image(story_paragraph), chatbot=chatbot, history=history, delay=0.)
# # 构建后续剧情引导
previously_on_story = ""
for s in self.story:
previously_on_story += s + '\n'
inputs_ = prompts_interact.format(previously_on_story=previously_on_story)
history_ = []
self.next_choices = yield from request_gpt_model_in_new_thread_with_ui_alive(
inputs_,
'请在以下几种故事走向中，选择一种。当然，您也可以给出您心中的其他故事走向。另外，如果您希望剧情立即收尾，请输入剧情走向，并以“剧情收尾”四个字提示程序。', self.llm_kwargs,
chatbot,
history_,
self.sys_prompt_
)
self.cur_task = 'user_choice'
elif self.cur_task == 'story_terminate':
"""
根据用户的提示，确定故事的结局
"""
previously_on_story = ""
for s in self.story:
previously_on_story += s + '\n'
inputs_ = prompts_terminate.format(previously_on_story=previously_on_story, user_choice=prompt)
history_ = []
story_paragraph = yield from request_gpt_model_in_new_thread_with_ui_alive(
inputs_, f'故事收尾（您的选择是：{prompt}）。', self.llm_kwargs,
chatbot, history_, self.sys_prompt_
)
# # 配图
yield from update_ui_lastest_msg(lastmsg=story_paragraph + '<br/>正在生成插图中 ...', chatbot=chatbot, history=history, delay=0.)
yield from update_ui_lastest_msg(lastmsg=story_paragraph + '<br/>'+ self.generate_story_image(story_paragraph), chatbot=chatbot, history=history, delay=0.)
# terminate game
self.delete_game = True
return

35
crazy_functions/game_fns/game_utils.py 普通文件
查看文件

@@ -0,0 +1,35 @@
from crazy_functions.json_fns.pydantic_io import GptJsonIO, JsonStringError
from request_llms.bridge_all import predict_no_ui_long_connection
def get_code_block(reply):
import re
pattern = r"```([\s\S]*?)```" # regex pattern to match code blocks
matches = re.findall(pattern, reply) # find all code blocks in text
if len(matches) == 1:
return "```" + matches[0] + "```" # code block
raise RuntimeError("GPT is not generating proper code.")
def is_same_thing(a, b, llm_kwargs):
from pydantic import BaseModel, Field
class IsSameThing(BaseModel):
is_same_thing: bool = Field(description="determine whether two objects are same thing.", default=False)
def run_gpt_fn(inputs, sys_prompt, history=[]):
return predict_no_ui_long_connection(
inputs=inputs, llm_kwargs=llm_kwargs,
history=history, sys_prompt=sys_prompt, observe_window=[]
)
gpt_json_io = GptJsonIO(IsSameThing)
inputs_01 = "Identity whether the user input and the target is the same thing: \n target object: {a} \n user input object: {b} \n\n\n".format(a=a, b=b)
inputs_01 += "\n\n\n Note that the user may describe the target object with a different language, e.g. cat and 猫 are the same thing."
analyze_res_cot_01 = run_gpt_fn(inputs_01, "", [])
inputs_02 = inputs_01 + gpt_json_io.format_instructions
analyze_res = run_gpt_fn(inputs_02, "", [inputs_01, analyze_res_cot_01])
try:
res = gpt_json_io.generate_output_auto_repair(analyze_res, run_gpt_fn)
return res.is_same_thing
except JsonStringError as e:
return False

37
crazy_functions/ipc_fns/mp.py 普通文件
查看文件

@@ -0,0 +1,37 @@
import platform
import pickle
import multiprocessing
def run_in_subprocess_wrapper_func(v_args):
func, args, kwargs, return_dict, exception_dict = pickle.loads(v_args)
import sys
try:
result = func(*args, **kwargs)
return_dict['result'] = result
except Exception as e:
exc_info = sys.exc_info()
exception_dict['exception'] = exc_info
def run_in_subprocess_with_timeout(func, timeout=60):
if platform.system() == 'Linux':
def wrapper(*args, **kwargs):
return_dict = multiprocessing.Manager().dict()
exception_dict = multiprocessing.Manager().dict()
v_args = pickle.dumps((func, args, kwargs, return_dict, exception_dict))
process = multiprocessing.Process(target=run_in_subprocess_wrapper_func, args=(v_args,))
process.start()
process.join(timeout)
if process.is_alive():
process.terminate()
raise TimeoutError(f'功能单元{str(func)}未能在规定时间内完成任务')
process.close()
if 'exception' in exception_dict:
# ooops, the subprocess ran into an exception
exc_info = exception_dict['exception']
raise exc_info[1].with_traceback(exc_info[2])
if 'result' in return_dict.keys():
# If the subprocess ran successfully, return the result
return return_dict['result']
return wrapper
else:
return func

125
crazy_functions/pdf_fns/breakdown_txt.py 普通文件
查看文件

@@ -0,0 +1,125 @@
from crazy_functions.ipc_fns.mp import run_in_subprocess_with_timeout
def force_breakdown(txt, limit, get_token_fn):
""" 当无法用标点、空行分割时，我们用最暴力的方法切割
"""
for i in reversed(range(len(txt))):
if get_token_fn(txt[:i]) < limit:
return txt[:i], txt[i:]
return "Tiktoken未知错误", "Tiktoken未知错误"
def maintain_storage(remain_txt_to_cut, remain_txt_to_cut_storage):
""" 为了加速计算，我们采样一个特殊的手段。当 remain_txt_to_cut > `_max` 时， 我们把 _max 后的文字转存至 remain_txt_to_cut_storage
当 remain_txt_to_cut < `_min` 时，我们再把 remain_txt_to_cut_storage 中的部分文字取出
"""
_min = int(5e4)
_max = int(1e5)
# print(len(remain_txt_to_cut), len(remain_txt_to_cut_storage))
if len(remain_txt_to_cut) < _min and len(remain_txt_to_cut_storage) > 0:
remain_txt_to_cut = remain_txt_to_cut + remain_txt_to_cut_storage
remain_txt_to_cut_storage = ""
if len(remain_txt_to_cut) > _max:
remain_txt_to_cut_storage = remain_txt_to_cut[_max:] + remain_txt_to_cut_storage
remain_txt_to_cut = remain_txt_to_cut[:_max]
return remain_txt_to_cut, remain_txt_to_cut_storage
def cut(limit, get_token_fn, txt_tocut, must_break_at_empty_line, break_anyway=False):
""" 文本切分
"""
res = []
total_len = len(txt_tocut)
fin_len = 0
remain_txt_to_cut = txt_tocut
remain_txt_to_cut_storage = ""
# 为了加速计算，我们采样一个特殊的手段。当 remain_txt_to_cut > `_max` 时， 我们把 _max 后的文字转存至 remain_txt_to_cut_storage
remain_txt_to_cut, remain_txt_to_cut_storage = maintain_storage(remain_txt_to_cut, remain_txt_to_cut_storage)
while True:
if get_token_fn(remain_txt_to_cut) <= limit:
# 如果剩余文本的token数小于限制，那么就不用切了
res.append(remain_txt_to_cut); fin_len+=len(remain_txt_to_cut)
break
else:
# 如果剩余文本的token数大于限制，那么就切
lines = remain_txt_to_cut.split('\n')
# 估计一个切分点
estimated_line_cut = limit / get_token_fn(remain_txt_to_cut) * len(lines)
estimated_line_cut = int(estimated_line_cut)
# 开始查找合适切分点的偏移（cnt）
cnt = 0
for cnt in reversed(range(estimated_line_cut)):
if must_break_at_empty_line:
# 首先尝试用双空行（\n\n）作为切分点
if lines[cnt] != "":
continue
prev = "\n".join(lines[:cnt])
post = "\n".join(lines[cnt:])
if get_token_fn(prev) < limit:
break
if cnt == 0:
# 如果没有找到合适的切分点
if break_anyway:
# 是否允许暴力切分
prev, post = force_breakdown(txt_tocut, limit, get_token_fn)
else:
# 不允许直接报错
raise RuntimeError(f"存在一行极长的文本！{txt_tocut}")
# 追加列表
res.append(prev); fin_len+=len(prev)
# 准备下一次迭代
remain_txt_to_cut = post
remain_txt_to_cut, remain_txt_to_cut_storage = maintain_storage(remain_txt_to_cut, remain_txt_to_cut_storage)
process = fin_len/total_len
print(f'正在文本切分 {int(process*100)}%')
if len(remain_txt_to_cut.strip()) == 0:
break
return res
def breakdown_text_to_satisfy_token_limit_(txt, limit, llm_model="gpt-3.5-turbo"):
""" 使用多种方式尝试切分文本，以满足 token 限制
"""
from request_llms.bridge_all import model_info
enc = model_info[llm_model]['tokenizer']
def get_token_fn(txt): return len(enc.encode(txt, disallowed_special=()))
try:
# 第1次尝试，将双空行（\n\n）作为切分点
return cut(limit, get_token_fn, txt, must_break_at_empty_line=True)
except RuntimeError:
try:
# 第2次尝试，将单空行（\n）作为切分点
return cut(limit, get_token_fn, txt, must_break_at_empty_line=False)
except RuntimeError:
try:
# 第3次尝试，将英文句号（.）作为切分点
res = cut(limit, get_token_fn, txt.replace('.', '。\n'), must_break_at_empty_line=False) # 这个中文的句号是故意的，作为一个标识而存在
return [r.replace('。\n', '.') for r in res]
except RuntimeError as e:
try:
# 第4次尝试，将中文句号（。）作为切分点
res = cut(limit, get_token_fn, txt.replace('。', '。。\n'), must_break_at_empty_line=False)
return [r.replace('。。\n', '。') for r in res]
except RuntimeError as e:
# 第5次尝试，没办法了，随便切一下吧
return cut(limit, get_token_fn, txt, must_break_at_empty_line=False, break_anyway=True)
breakdown_text_to_satisfy_token_limit = run_in_subprocess_with_timeout(breakdown_text_to_satisfy_token_limit_, timeout=60)
if __name__ == '__main__':
from crazy_functions.crazy_utils import read_and_clean_pdf_text
file_content, page_one = read_and_clean_pdf_text("build/assets/at.pdf")
from request_llms.bridge_all import model_info
for i in range(5):
file_content += file_content
print(len(file_content))
TOKEN_LIMIT_PER_FRAGMENT = 2500
res = breakdown_text_to_satisfy_token_limit(file_content, TOKEN_LIMIT_PER_FRAGMENT)

0
crazy_functions/vector_fns/__init__.py 普通文件
查看文件

70
crazy_functions/vector_fns/general_file_loader.py 普通文件
查看文件

@@ -0,0 +1,70 @@
# From project chatglm-langchain
from langchain.document_loaders import UnstructuredFileLoader
from langchain.text_splitter import CharacterTextSplitter
import re
from typing import List
class ChineseTextSplitter(CharacterTextSplitter):
def __init__(self, pdf: bool = False, sentence_size: int = None, **kwargs):
super().__init__(**kwargs)
self.pdf = pdf
self.sentence_size = sentence_size
def split_text1(self, text: str) -> List[str]:
if self.pdf:
text = re.sub(r"\n{3,}", "\n", text)
text = re.sub('\s', ' ', text)
text = text.replace("\n\n", "")
sent_sep_pattern = re.compile('([﹒﹔﹖﹗．。！？]["’”」』]{0,2}|(?=["‘“「『]{1,2}|$))') # del ：；
sent_list = []
for ele in sent_sep_pattern.split(text):
if sent_sep_pattern.match(ele) and sent_list:
sent_list[-1] += ele
elif ele:
sent_list.append(ele)
return sent_list
def split_text(self, text: str) -> List[str]: ##此处需要进一步优化逻辑
if self.pdf:
text = re.sub(r"\n{3,}", r"\n", text)
text = re.sub('\s', " ", text)
text = re.sub("\n\n", "", text)
text = re.sub(r'([;；.!?。！？\?])([^”’])', r"\1\n\2", text) # 单字符断句符
text = re.sub(r'(\.{6})([^"’”」』])', r"\1\n\2", text) # 英文省略号
text = re.sub(r'(\…{2})([^"’”」』])', r"\1\n\2", text) # 中文省略号
text = re.sub(r'([;；!?。！？\?]["’”」』]{0,2})([^;；!?，。！？\?])', r'\1\n\2', text)
# 如果双引号前有终止符，那么双引号才是句子的终点，把分句符\n放到双引号后，注意前面的几句都小心保留了双引号
text = text.rstrip() # 段尾如果有多余的\n就去掉它
# 很多规则中会考虑分号;，但是这里我把它忽略不计，破折号、英文双引号等同样忽略，需要的再做些简单调整即可。
ls = [i for i in text.split("\n") if i]
for ele in ls:
if len(ele) > self.sentence_size:
ele1 = re.sub(r'([,，.]["’”」』]{0,2})([^,，.])', r'\1\n\2', ele)
ele1_ls = ele1.split("\n")
for ele_ele1 in ele1_ls:
if len(ele_ele1) > self.sentence_size:
ele_ele2 = re.sub(r'([\n]{1,}| {2,}["’”」』]{0,2})([^\s])', r'\1\n\2', ele_ele1)
ele2_ls = ele_ele2.split("\n")
for ele_ele2 in ele2_ls:
if len(ele_ele2) > self.sentence_size:
ele_ele3 = re.sub('( ["’”」』]{0,2})([^ ])', r'\1\n\2', ele_ele2)
ele2_id = ele2_ls.index(ele_ele2)
ele2_ls = ele2_ls[:ele2_id] + [i for i in ele_ele3.split("\n") if i] + ele2_ls[
ele2_id + 1:]
ele_id = ele1_ls.index(ele_ele1)
ele1_ls = ele1_ls[:ele_id] + [i for i in ele2_ls if i] + ele1_ls[ele_id + 1:]
id = ls.index(ele)
ls = ls[:id] + [i for i in ele1_ls if i] + ls[id + 1:]
return ls
def load_file(filepath, sentence_size):
loader = UnstructuredFileLoader(filepath, mode="elements")
textsplitter = ChineseTextSplitter(pdf=False, sentence_size=sentence_size)
docs = loader.load_and_split(text_splitter=textsplitter)
# write_check_file(filepath, docs)
return docs

338
crazy_functions/vector_fns/vector_database.py 普通文件
查看文件

@@ -0,0 +1,338 @@
# From project chatglm-langchain
import threading
from toolbox import Singleton
import os
import shutil
import os
import uuid
import tqdm
from langchain.vectorstores import FAISS
from langchain.docstore.document import Document
from typing import List, Tuple
import numpy as np
from crazy_functions.vector_fns.general_file_loader import load_file
embedding_model_dict = {
"ernie-tiny": "nghuyong/ernie-3.0-nano-zh",
"ernie-base": "nghuyong/ernie-3.0-base-zh",
"text2vec-base": "shibing624/text2vec-base-chinese",
"text2vec": "GanymedeNil/text2vec-large-chinese",
}
# Embedding model name
EMBEDDING_MODEL = "text2vec"
# Embedding running device
EMBEDDING_DEVICE = "cpu"
# 基于上下文的prompt模版，请务必保留"{question}"和"{context}"
PROMPT_TEMPLATE = """已知信息：
{context}
根据上述已知信息，简洁和专业的来回答用户的问题。如果无法从中得到答案，请说 “根据已知信息无法回答该问题” 或 “没有提供足够的相关信息”，不允许在答案中添加编造成分，答案请使用中文。 问题是：{question}"""
# 文本分句长度
SENTENCE_SIZE = 100
# 匹配后单段上下文长度
CHUNK_SIZE = 250
# LLM input history length
LLM_HISTORY_LEN = 3
# return top-k text chunk from vector store
VECTOR_SEARCH_TOP_K = 5
# 知识检索内容相关度 Score, 数值范围约为0-1100，如果为0，则不生效，经测试设置为小于500时，匹配结果更精准
VECTOR_SEARCH_SCORE_THRESHOLD = 0
NLTK_DATA_PATH = os.path.join(os.path.dirname(os.path.dirname(__file__)), "nltk_data")
FLAG_USER_NAME = uuid.uuid4().hex
# 是否开启跨域，默认为False，如果需要开启，请设置为True
# is open cross domain
OPEN_CROSS_DOMAIN = False
def similarity_search_with_score_by_vector(
self, embedding: List[float], k: int = 4
) -> List[Tuple[Document, float]]:
def seperate_list(ls: List[int]) -> List[List[int]]:
lists = []
ls1 = [ls[0]]
for i in range(1, len(ls)):
if ls[i - 1] + 1 == ls[i]:
ls1.append(ls[i])
else:
lists.append(ls1)
ls1 = [ls[i]]
lists.append(ls1)
return lists
scores, indices = self.index.search(np.array([embedding], dtype=np.float32), k)
docs = []
id_set = set()
store_len = len(self.index_to_docstore_id)
for j, i in enumerate(indices[0]):
if i == -1 or 0 < self.score_threshold < scores[0][j]:
# This happens when not enough docs are returned.
continue
_id = self.index_to_docstore_id[i]
doc = self.docstore.search(_id)
if not self.chunk_conent:
if not isinstance(doc, Document):
raise ValueError(f"Could not find document for id {_id}, got {doc}")
doc.metadata["score"] = int(scores[0][j])
docs.append(doc)
continue
id_set.add(i)
docs_len = len(doc.page_content)
for k in range(1, max(i, store_len - i)):
break_flag = False
for l in [i + k, i - k]:
if 0 <= l < len(self.index_to_docstore_id):
_id0 = self.index_to_docstore_id[l]
doc0 = self.docstore.search(_id0)
if docs_len + len(doc0.page_content) > self.chunk_size:
break_flag = True
break
elif doc0.metadata["source"] == doc.metadata["source"]:
docs_len += len(doc0.page_content)
id_set.add(l)
if break_flag:
break
if not self.chunk_conent:
return docs
if len(id_set) == 0 and self.score_threshold > 0:
return []
id_list = sorted(list(id_set))
id_lists = seperate_list(id_list)
for id_seq in id_lists:
for id in id_seq:
if id == id_seq[0]:
_id = self.index_to_docstore_id[id]
doc = self.docstore.search(_id)
else:
_id0 = self.index_to_docstore_id[id]
doc0 = self.docstore.search(_id0)
doc.page_content += " " + doc0.page_content
if not isinstance(doc, Document):
raise ValueError(f"Could not find document for id {_id}, got {doc}")
doc_score = min([scores[0][id] for id in [indices[0].tolist().index(i) for i in id_seq if i in indices[0]]])
doc.metadata["score"] = int(doc_score)
docs.append(doc)
return docs
class LocalDocQA:
llm: object = None
embeddings: object = None
top_k: int = VECTOR_SEARCH_TOP_K
chunk_size: int = CHUNK_SIZE
chunk_conent: bool = True
score_threshold: int = VECTOR_SEARCH_SCORE_THRESHOLD
def init_cfg(self,
top_k=VECTOR_SEARCH_TOP_K,
):
self.llm = None
self.top_k = top_k
def init_knowledge_vector_store(self,
filepath,
vs_path: str or os.PathLike = None,
sentence_size=SENTENCE_SIZE,
text2vec=None):
loaded_files = []
failed_files = []
if isinstance(filepath, str):
if not os.path.exists(filepath):
print("路径不存在")
return None
elif os.path.isfile(filepath):
file = os.path.split(filepath)[-1]
try:
docs = load_file(filepath, SENTENCE_SIZE)
print(f"{file} 已成功加载")
loaded_files.append(filepath)
except Exception as e:
print(e)
print(f"{file} 未能成功加载")
return None
elif os.path.isdir(filepath):
docs = []
for file in tqdm(os.listdir(filepath), desc="加载文件"):
fullfilepath = os.path.join(filepath, file)
try:
docs += load_file(fullfilepath, SENTENCE_SIZE)
loaded_files.append(fullfilepath)
except Exception as e:
print(e)
failed_files.append(file)
if len(failed_files) > 0:
print("以下文件未能成功加载：")
for file in failed_files:
print(f"{file}\n")
else:
docs = []
for file in filepath:
docs += load_file(file, SENTENCE_SIZE)
print(f"{file} 已成功加载")
loaded_files.append(file)
if len(docs) > 0:
print("文件加载完毕，正在生成向量库")
if vs_path and os.path.isdir(vs_path):
try:
self.vector_store = FAISS.load_local(vs_path, text2vec)
self.vector_store.add_documents(docs)
except:
self.vector_store = FAISS.from_documents(docs, text2vec)
else:
self.vector_store = FAISS.from_documents(docs, text2vec) # docs 为Document列表
self.vector_store.save_local(vs_path)
return vs_path, loaded_files
else:
raise RuntimeError("文件加载失败，请检查文件格式是否正确")
def get_loaded_file(self, vs_path):
ds = self.vector_store.docstore
return set([ds._dict[k].metadata['source'].split(vs_path)[-1] for k in ds._dict])
# query 查询内容
# vs_path 知识库路径
# chunk_conent 是否启用上下文关联
# score_threshold 搜索匹配score阈值
# vector_search_top_k 搜索知识库内容条数，默认搜索5条结果
# chunk_sizes 匹配单段内容的连接上下文长度
def get_knowledge_based_conent_test(self, query, vs_path, chunk_conent,
score_threshold=VECTOR_SEARCH_SCORE_THRESHOLD,
vector_search_top_k=VECTOR_SEARCH_TOP_K, chunk_size=CHUNK_SIZE,
text2vec=None):
self.vector_store = FAISS.load_local(vs_path, text2vec)
self.vector_store.chunk_conent = chunk_conent
self.vector_store.score_threshold = score_threshold
self.vector_store.chunk_size = chunk_size
embedding = self.vector_store.embedding_function.embed_query(query)
related_docs_with_score = similarity_search_with_score_by_vector(self.vector_store, embedding, k=vector_search_top_k)
if not related_docs_with_score:
response = {"query": query,
"source_documents": []}
return response, ""
# prompt = f"{query}. You should answer this question using information from following documents: \n\n"
prompt = f"{query}. 你必须利用以下文档中包含的信息回答这个问题: \n\n---\n\n"
prompt += "\n\n".join([f"({k}): " + doc.page_content for k, doc in enumerate(related_docs_with_score)])
prompt += "\n\n---\n\n"
prompt = prompt.encode('utf-8', 'ignore').decode() # avoid reading non-utf8 chars
# print(prompt)
response = {"query": query, "source_documents": related_docs_with_score}
return response, prompt
def construct_vector_store(vs_id, vs_path, files, sentence_size, history, one_conent, one_content_segmentation, text2vec):
for file in files:
assert os.path.exists(file), "输入文件不存在：" + file
import nltk
if NLTK_DATA_PATH not in nltk.data.path: nltk.data.path = [NLTK_DATA_PATH] + nltk.data.path
local_doc_qa = LocalDocQA()
local_doc_qa.init_cfg()
filelist = []
if not os.path.exists(os.path.join(vs_path, vs_id)):
os.makedirs(os.path.join(vs_path, vs_id))
for file in files:
file_name = file.name if not isinstance(file, str) else file
filename = os.path.split(file_name)[-1]
shutil.copyfile(file_name, os.path.join(vs_path, vs_id, filename))
filelist.append(os.path.join(vs_path, vs_id, filename))
vs_path, loaded_files = local_doc_qa.init_knowledge_vector_store(filelist, os.path.join(vs_path, vs_id), sentence_size, text2vec)
if len(loaded_files):
file_status = f"已添加 {'、'.join([os.path.split(i)[-1] for i in loaded_files if i])} 内容至知识库，并已加载知识库，请开始提问"
else:
pass
# file_status = "文件未成功加载，请重新上传文件"
# print(file_status)
return local_doc_qa, vs_path
@Singleton
class knowledge_archive_interface():
def __init__(self) -> None:
self.threadLock = threading.Lock()
self.current_id = ""
self.kai_path = None
self.qa_handle = None
self.text2vec_large_chinese = None
def get_chinese_text2vec(self):
if self.text2vec_large_chinese is None:
# < -------------------预热文本向量化模组--------------- >
from toolbox import ProxyNetworkActivate
print('Checking Text2vec ...')
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
with ProxyNetworkActivate('Download_LLM'): # 临时地激活代理网络
self.text2vec_large_chinese = HuggingFaceEmbeddings(model_name="GanymedeNil/text2vec-large-chinese")
return self.text2vec_large_chinese
def feed_archive(self, file_manifest, vs_path, id="default"):
self.threadLock.acquire()
# import uuid
self.current_id = id
self.qa_handle, self.kai_path = construct_vector_store(
vs_id=self.current_id,
vs_path=vs_path,
files=file_manifest,
sentence_size=100,
history=[],
one_conent="",
one_content_segmentation="",
text2vec = self.get_chinese_text2vec(),
)
self.threadLock.release()
def get_current_archive_id(self):
return self.current_id
def get_loaded_file(self, vs_path):
return self.qa_handle.get_loaded_file(vs_path)
def answer_with_archive_by_id(self, txt, id, vs_path):
self.threadLock.acquire()
if not self.current_id == id:
self.current_id = id
self.qa_handle, self.kai_path = construct_vector_store(
vs_id=self.current_id,
vs_path=vs_path,
files=[],
sentence_size=100,
history=[],
one_conent="",
one_content_segmentation="",
text2vec = self.get_chinese_text2vec(),
)
VECTOR_SEARCH_SCORE_THRESHOLD = 0
VECTOR_SEARCH_TOP_K = 4
CHUNK_SIZE = 512
resp, prompt = self.qa_handle.get_knowledge_based_conent_test(
query = txt,
vs_path = self.kai_path,
score_threshold=VECTOR_SEARCH_SCORE_THRESHOLD,
vector_search_top_k=VECTOR_SEARCH_TOP_K,
chunk_conent=True,
chunk_size=CHUNK_SIZE,
text2vec = self.get_chinese_text2vec(),
)
self.threadLock.release()
return resp, prompt

40
crazy_functions/互动小游戏.py 普通文件
查看文件

@@ -0,0 +1,40 @@
from toolbox import CatchException, update_ui, update_ui_lastest_msg
from crazy_functions.multi_stage.multi_stage_utils import GptAcademicGameBaseState
from crazy_functions.crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
from request_llms.bridge_all import predict_no_ui_long_connection
from crazy_functions.game_fns.game_utils import get_code_block, is_same_thing
@CatchException
def 随机小游戏(prompt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
from crazy_functions.game_fns.game_interactive_story import MiniGame_ResumeStory
# 清空历史
history = []
# 选择游戏
cls = MiniGame_ResumeStory
# 如果之前已经初始化了游戏实例，则继续该实例；否则重新初始化
state = cls.sync_state(chatbot,
llm_kwargs,
cls,
plugin_name='MiniGame_ResumeStory',
callback_fn='crazy_functions.互动小游戏->随机小游戏',
lock_plugin=True
)
yield from state.continue_game(prompt, chatbot, history)
@CatchException
def 随机小游戏1(prompt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
from crazy_functions.game_fns.game_ascii_art import MiniGame_ASCII_Art
# 清空历史
history = []
# 选择游戏
cls = MiniGame_ASCII_Art
# 如果之前已经初始化了游戏实例，则继续该实例；否则重新初始化
state = cls.sync_state(chatbot,
llm_kwargs,
cls,
plugin_name='MiniGame_ASCII_Art',
callback_fn='crazy_functions.互动小游戏->随机小游戏1',
lock_plugin=True
)
yield from state.continue_game(prompt, chatbot, history)

117
crazy_functions/知识库问答.py 普通文件
查看文件

@@ -0,0 +1,117 @@
from toolbox import CatchException, update_ui, ProxyNetworkActivate, update_ui_lastest_msg, get_log_folder, get_user
from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive, get_files_from_everything
install_msg ="""
1. python -m pip install torch --index-url https://download.pytorch.org/whl/cpu
2. python -m pip install transformers protobuf langchain sentence-transformers faiss-cpu nltk beautifulsoup4 bitsandbytes tabulate icetk --upgrade
3. python -m pip install unstructured[all-docs] --upgrade
4. python -c 'import nltk; nltk.download("punkt")'
"""
@CatchException
def 知识库文件注入(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，暂时没有用武之地
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
"""
history = [] # 清空历史，以免输入溢出
# < --------------------读取参数--------------- >
if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg")
kai_id = plugin_kwargs.get("advanced_arg", 'default')
chatbot.append((f"向`{kai_id}`知识库中添加文件。", "[Local Message] 从一批文件(txt, md, tex)中读取数据构建知识库, 然后进行问答。"))
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
# resolve deps
try:
# from zh_langchain import construct_vector_store
# from langchain.embeddings.huggingface import HuggingFaceEmbeddings
from crazy_functions.vector_fns.vector_database import knowledge_archive_interface
except Exception as e:
chatbot.append(["依赖不足", f"{str(e)}\n\n导入依赖失败。请用以下命令安装" + install_msg])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
# from .crazy_utils import try_install_deps
# try_install_deps(['zh_langchain==0.2.1', 'pypinyin'], reload_m=['pypinyin', 'zh_langchain'])
# yield from update_ui_lastest_msg("安装完成，您可以再次重试。", chatbot, history)
return
# < --------------------读取文件--------------- >
file_manifest = []
spl = ["txt", "doc", "docx", "email", "epub", "html", "json", "md", "msg", "pdf", "ppt", "pptx", "rtf"]
for sp in spl:
_, file_manifest_tmp, _ = get_files_from_everything(txt, type=f'.{sp}')
file_manifest += file_manifest_tmp
if len(file_manifest) == 0:
chatbot.append(["没有找到任何可读取文件", "当前支持的格式包括: txt, md, docx, pptx, pdf, json等"])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
return
# < -------------------预热文本向量化模组--------------- >
chatbot.append(['<br/>'.join(file_manifest), "正在预热文本向量化模组, 如果是第一次运行, 将消耗较长时间下载中文向量化模型..."])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
print('Checking Text2vec ...')
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
with ProxyNetworkActivate('Download_LLM'): # 临时地激活代理网络
HuggingFaceEmbeddings(model_name="GanymedeNil/text2vec-large-chinese")
# < -------------------构建知识库--------------- >
chatbot.append(['<br/>'.join(file_manifest), "正在构建知识库..."])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
print('Establishing knowledge archive ...')
with ProxyNetworkActivate('Download_LLM'): # 临时地激活代理网络
kai = knowledge_archive_interface()
vs_path = get_log_folder(user=get_user(chatbot), plugin_name='vec_store')
kai.feed_archive(file_manifest=file_manifest, vs_path=vs_path, id=kai_id)
kai_files = kai.get_loaded_file(vs_path=vs_path)
kai_files = '<br/>'.join(kai_files)
# chatbot.append(['知识库构建成功', "正在将知识库存储至cookie中"])
# yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
# chatbot._cookies['langchain_plugin_embedding'] = kai.get_current_archive_id()
# chatbot._cookies['lock_plugin'] = 'crazy_functions.知识库文件注入->读取知识库作答'
# chatbot.append(['完成', "“根据知识库作答”函数插件已经接管问答系统, 提问吧! 但注意, 您接下来不能再使用其他插件了，刷新页面即可以退出知识库问答模式。"])
chatbot.append(['构建完成', f"当前知识库内的有效文件：\n\n---\n\n{kai_files}\n\n---\n\n请切换至“知识库问答”插件进行知识库访问, 或者使用此插件继续上传更多文件。"])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # 由于请求gpt需要一段时间，我们先及时地做一次界面更新
@CatchException
def 读取知识库作答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port=-1):
# resolve deps
try:
# from zh_langchain import construct_vector_store
# from langchain.embeddings.huggingface import HuggingFaceEmbeddings
from crazy_functions.vector_fns.vector_database import knowledge_archive_interface
except Exception as e:
chatbot.append(["依赖不足", f"{str(e)}\n\n导入依赖失败。请用以下命令安装" + install_msg])
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
# from .crazy_utils import try_install_deps
# try_install_deps(['zh_langchain==0.2.1', 'pypinyin'], reload_m=['pypinyin', 'zh_langchain'])
# yield from update_ui_lastest_msg("安装完成，您可以再次重试。", chatbot, history)
return
# < ------------------- --------------- >
kai = knowledge_archive_interface()
if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg")
kai_id = plugin_kwargs.get("advanced_arg", 'default')
vs_path = get_log_folder(user=get_user(chatbot), plugin_name='vec_store')
resp, prompt = kai.answer_with_archive_by_id(txt, kai_id, vs_path)
chatbot.append((txt, f'[知识库 {kai_id}] ' + prompt))
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # 由于请求gpt需要一段时间，我们先及时地做一次界面更新
gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
inputs=prompt, inputs_show_user=txt,
llm_kwargs=llm_kwargs, chatbot=chatbot, history=[],
sys_prompt=system_prompt
)
history.extend((prompt, gpt_say))
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # 由于请求gpt需要一段时间，我们先及时地做一次界面更新

53
docs/GithubAction+AllCapacityBeta 普通文件
查看文件

@@ -0,0 +1,53 @@
# docker build -t gpt-academic-all-capacity -f docs/GithubAction+AllCapacity --network=host --build-arg http_proxy=http://localhost:10881 --build-arg https_proxy=http://localhost:10881 .
# docker build -t gpt-academic-all-capacity -f docs/GithubAction+AllCapacityBeta --network=host .
# docker run -it --net=host gpt-academic-all-capacity bash
# 从NVIDIA源，从而支持显卡（检查宿主的nvidia-smi中的cuda版本必须>=11.3）
FROM fuqingxu/11.3.1-runtime-ubuntu20.04-with-texlive:latest
# use python3 as the system default python
WORKDIR /gpt
RUN curl -sS https://bootstrap.pypa.io/get-pip.py | python3.8
# # 非必要步骤，更换pip源 （以下三行，可以删除）
# RUN echo '[global]' > /etc/pip.conf && \
# echo 'index-url = https://mirrors.aliyun.com/pypi/simple/' >> /etc/pip.conf && \
# echo 'trusted-host = mirrors.aliyun.com' >> /etc/pip.conf
# 下载pytorch
RUN python3 -m pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu113
# 准备pip依赖
RUN python3 -m pip install openai numpy arxiv rich
RUN python3 -m pip install colorama Markdown pygments pymupdf
RUN python3 -m pip install python-docx moviepy pdfminer
RUN python3 -m pip install zh_langchain==0.2.1 pypinyin
RUN python3 -m pip install rarfile py7zr
RUN python3 -m pip install aliyun-python-sdk-core==2.13.3 pyOpenSSL webrtcvad scipy git+https://github.com/aliyun/alibabacloud-nls-python-sdk.git
# 下载分支
WORKDIR /gpt
RUN git clone --depth=1 https://github.com/binary-husky/gpt_academic.git
WORKDIR /gpt/gpt_academic
RUN git clone --depth=1 https://github.com/OpenLMLab/MOSS.git request_llms/moss
RUN python3 -m pip install -r requirements.txt
RUN python3 -m pip install -r request_llms/requirements_moss.txt
RUN python3 -m pip install -r request_llms/requirements_qwen.txt
RUN python3 -m pip install -r request_llms/requirements_chatglm.txt
RUN python3 -m pip install -r request_llms/requirements_newbing.txt
RUN python3 -m pip install nougat-ocr
# 预热Tiktoken模块
RUN python3 -c 'from check_proxy import warm_up_modules; warm_up_modules()'
# 安装知识库插件的额外依赖
RUN apt-get update && apt-get install libgl1 -y
RUN pip3 install transformers protobuf langchain sentence-transformers faiss-cpu nltk beautifulsoup4 bitsandbytes tabulate icetk --upgrade
RUN pip3 install unstructured[all-docs] --upgrade
RUN python3 -c 'from check_proxy import warm_up_vectordb; warm_up_vectordb()'
RUN rm -rf /usr/local/lib/python3.8/dist-packages/tests
# COPY .cache /root/.cache
# COPY config_private.py config_private.py
# 启动
CMD ["python3", "-u", "main.py"]

26
docs/GithubAction+NoLocal+Vectordb 普通文件
查看文件

@@ -0,0 +1,26 @@
# 此Dockerfile适用于“无本地模型”的环境构建，如果需要使用chatglm等本地模型，请参考 docs/Dockerfile+ChatGLM
# 如何构建: 先修改 `config.py`， 然后 docker build -t gpt-academic-nolocal-vs -f docs/GithubAction+NoLocal+Vectordb .
# 如何运行: docker run --rm -it --net=host gpt-academic-nolocal-vs
FROM python:3.11
# 指定路径
WORKDIR /gpt
# 装载项目文件
COPY . .
# 安装依赖
RUN pip3 install -r requirements.txt
# 安装知识库插件的额外依赖
RUN apt-get update && apt-get install libgl1 -y
RUN pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cpu
RUN pip3 install transformers protobuf langchain sentence-transformers faiss-cpu nltk beautifulsoup4 bitsandbytes tabulate icetk --upgrade
RUN pip3 install unstructured[all-docs] --upgrade
RUN python3 -c 'from check_proxy import warm_up_vectordb; warm_up_vectordb()'
# 可选步骤，用于预热模块
RUN python3 -c 'from check_proxy import warm_up_modules; warm_up_modules()'
# 启动
CMD ["python3", "-u", "main.py"]

59
request_llms/bridge_qwen_local.py 普通文件
查看文件

@@ -0,0 +1,59 @@
model_name = "Qwen_Local"
cmd_to_install = "`pip install -r request_llms/requirements_qwen_local.txt`"
from toolbox import ProxyNetworkActivate, get_conf
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns
# ------------------------------------------------------------------------------------------------------------------------
# 🔌💻 Local Model
# ------------------------------------------------------------------------------------------------------------------------
class GetQwenLMHandle(LocalLLMHandle):
def load_model_info(self):
# 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行
self.model_name = model_name
self.cmd_to_install = cmd_to_install
def load_model_and_tokenizer(self):
# 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行
# from modelscope import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
from transformers import AutoModelForCausalLM, AutoTokenizer
from transformers.generation import GenerationConfig
with ProxyNetworkActivate('Download_LLM'):
model_id = get_conf('QWEN_LOCAL_MODEL_SELECTION')
self._tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True, resume_download=True)
# use fp16
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True).eval()
model.generation_config = GenerationConfig.from_pretrained(model_id, trust_remote_code=True) # 可指定不同的生成长度、top_p等相关超参
self._model = model
return self._model, self._tokenizer
def llm_stream_generator(self, **kwargs):
# 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行
def adaptor(kwargs):
query = kwargs['query']
max_length = kwargs['max_length']
top_p = kwargs['top_p']
temperature = kwargs['temperature']
history = kwargs['history']
return query, max_length, top_p, temperature, history
query, max_length, top_p, temperature, history = adaptor(kwargs)
for response in self._model.chat_stream(self._tokenizer, query, history=history):
yield response
def try_to_import_special_deps(self, **kwargs):
# import something that will raise error if the user does not install requirement_*.txt
# 🏃‍♂️🏃‍♂️🏃‍♂️ 主进程执行
import importlib
importlib.import_module('modelscope')
# ------------------------------------------------------------------------------------------------------------------------
# 🔌💻 GPT-Academic Interface
# ------------------------------------------------------------------------------------------------------------------------
predict_no_ui_long_connection, predict = get_local_llm_predict_fns(GetQwenLMHandle, model_name)

94
request_llms/com_qwenapi.py 普通文件
查看文件

@@ -0,0 +1,94 @@
from http import HTTPStatus
from toolbox import get_conf
import threading
import logging
timeout_bot_msg = '[Local Message] Request timeout. Network error.'
class QwenRequestInstance():
def __init__(self):
import dashscope
self.time_to_yield_event = threading.Event()
self.time_to_exit_event = threading.Event()
self.result_buf = ""
def validate_key():
DASHSCOPE_API_KEY = get_conf("DASHSCOPE_API_KEY")
if DASHSCOPE_API_KEY == '': return False
return True
if not validate_key():
raise RuntimeError('请配置 DASHSCOPE_API_KEY')
dashscope.api_key = get_conf("DASHSCOPE_API_KEY")
def generate(self, inputs, llm_kwargs, history, system_prompt):
# import _thread as thread
from dashscope import Generation
QWEN_MODEL = {
'qwen-turbo': Generation.Models.qwen_turbo,
'qwen-plus': Generation.Models.qwen_plus,
'qwen-max': Generation.Models.qwen_max,
}[llm_kwargs['llm_model']]
top_p = llm_kwargs.get('top_p', 0.8)
if top_p == 0: top_p += 1e-5
if top_p == 1: top_p -= 1e-5
self.result_buf = ""
responses = Generation.call(
model=QWEN_MODEL,
messages=generate_message_payload(inputs, llm_kwargs, history, system_prompt),
top_p=top_p,
temperature=llm_kwargs.get('temperature', 1.0),
result_format='message',
stream=True,
incremental_output=True
)
for response in responses:
if response.status_code == HTTPStatus.OK:
if response.output.choices[0].finish_reason == 'stop':
yield self.result_buf
break
elif response.output.choices[0].finish_reason == 'length':
self.result_buf += "[Local Message] 生成长度过长，后续输出被截断"
yield self.result_buf
break
else:
self.result_buf += response.output.choices[0].message.content
yield self.result_buf
else:
self.result_buf += f"[Local Message] 请求错误：状态码：{response.status_code}，错误码:{response.code}，消息：{response.message}"
yield self.result_buf
break
logging.info(f'[raw_input] {inputs}')
logging.info(f'[response] {self.result_buf}')
return self.result_buf
def generate_message_payload(inputs, llm_kwargs, history, system_prompt):
conversation_cnt = len(history) // 2
if system_prompt == '': system_prompt = 'Hello!'
messages = [{"role": "user", "content": system_prompt}, {"role": "assistant", "content": "Certainly!"}]
if conversation_cnt:
for index in range(0, 2*conversation_cnt, 2):
what_i_have_asked = {}
what_i_have_asked["role"] = "user"
what_i_have_asked["content"] = history[index]
what_gpt_answer = {}
what_gpt_answer["role"] = "assistant"
what_gpt_answer["content"] = history[index+1]
if what_i_have_asked["content"] != "":
if what_gpt_answer["content"] == "":
continue
if what_gpt_answer["content"] == timeout_bot_msg:
continue
messages.append(what_i_have_asked)
messages.append(what_gpt_answer)
else:
messages[-1]['content'] = what_gpt_answer['content']
what_i_ask_now = {}
what_i_ask_now["role"] = "user"
what_i_ask_now["content"] = inputs
messages.append(what_i_ask_now)
return messages

5
request_llms/requirements_qwen_local.txt 普通文件
查看文件

@@ -0,0 +1,5 @@
modelscope
transformers_stream_generator
auto-gptq
optimum
urllib3<2

17
tests/test_vector_plugins.py 普通文件
查看文件

@@ -0,0 +1,17 @@
"""
对项目中的各个插件进行测试。运行方法：直接运行 python tests/test_plugins.py
"""
import os, sys
def validate_path(): dir_name = os.path.dirname(__file__); root_dir_assume = os.path.abspath(dir_name + '/..'); os.chdir(root_dir_assume); sys.path.append(root_dir_assume)
validate_path() # 返回项目根路径
if __name__ == "__main__":
from tests.test_utils import plugin_test
plugin_test(plugin='crazy_functions.知识库问答->知识库文件注入', main_input="./README.md")
plugin_test(plugin='crazy_functions.知识库问答->读取知识库作答', main_input="What is the installation method？")
plugin_test(plugin='crazy_functions.知识库问答->读取知识库作答', main_input="远程云服务器部署？")

0
themes/cookies.py 普通文件
查看文件