提升稳定性
This commit is contained in:
qingxu fu
parent
91d07c329a
commit
e6cf5532a9
@ -29,7 +29,7 @@ def get_crazy_functions():
|
||||
"Color": "stop", # 按钮颜色
|
||||
"Function": HotReload(解析一个C项目的头文件)
|
||||
},
|
||||
"解析整个C++项目（.cpp/.h）": {
|
||||
"解析整个C++项目（.cpp/.hpp/.c/.h）": {
|
||||
"Color": "stop", # 按钮颜色
|
||||
"AsButton": False, # 加入下拉菜单中
|
||||
"Function": HotReload(解析一个C项目)
|
||||
|
||||
70
crazy_functions/Latex全文润色.py
Normal file
70
crazy_functions/Latex全文润色.py
Normal file
@ -0,0 +1,70 @@
|
||||
from request_llm.bridge_chatgpt import predict_no_ui
|
||||
from toolbox import CatchException, report_execption, write_results_to_file, predict_no_ui_but_counting_down
|
||||
fast_debug = False
|
||||
|
||||
|
||||
def 解析Paper(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt):
|
||||
import time, glob, os
|
||||
print('begin analysis on:', file_manifest)
|
||||
for index, fp in enumerate(file_manifest):
|
||||
with open(fp, 'r', encoding='utf-8') as f:
|
||||
file_content = f.read()
|
||||
|
||||
prefix = "接下来请你逐文件分析下面的论文文件，概括其内容" if index==0 else ""
|
||||
i_say = prefix + f'请对下面的文章片段用中文做一个概述，文件名是{os.path.relpath(fp, project_folder)}，文章内容是 ```{file_content}```'
|
||||
i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的文章片段做一个概述: {os.path.abspath(fp)}'
|
||||
chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
|
||||
print('[1] yield chatbot, history')
|
||||
yield chatbot, history, '正常'
|
||||
|
||||
if not fast_debug:
|
||||
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
|
||||
|
||||
print('[2] end gpt req')
|
||||
chatbot[-1] = (i_say_show_user, gpt_say)
|
||||
history.append(i_say_show_user); history.append(gpt_say)
|
||||
print('[3] yield chatbot, history')
|
||||
yield chatbot, history, msg
|
||||
print('[4] next')
|
||||
if not fast_debug: time.sleep(2)
|
||||
|
||||
all_file = ', '.join([os.path.relpath(fp, project_folder) for index, fp in enumerate(file_manifest)])
|
||||
i_say = f'根据以上你自己的分析，对全文进行概括，用学术性语言写一段中文摘要，然后再写一段英文摘要（包括{all_file}）。'
|
||||
chatbot.append((i_say, "[Local Message] waiting gpt response."))
|
||||
yield chatbot, history, '正常'
|
||||
|
||||
if not fast_debug:
|
||||
msg = '正常'
|
||||
# ** gpt request **
|
||||
gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say, chatbot, top_p, temperature, history=history) # 带超时倒计时
|
||||
|
||||
chatbot[-1] = (i_say, gpt_say)
|
||||
history.append(i_say); history.append(gpt_say)
|
||||
yield chatbot, history, msg
|
||||
res = write_results_to_file(history)
|
||||
chatbot.append(("完成了吗？", res))
|
||||
yield chatbot, history, msg
|
||||
|
||||
|
||||
|
||||
@CatchException
|
||||
def 读文章写摘要(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}/**/*.tex', recursive=True)] # + \
|
||||
# [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)] + \
|
||||
# [f for f in glob.glob(f'{project_folder}/**/*.c', 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.tex文件: {txt}")
|
||||
yield chatbot, history, '正常'
|
||||
return
|
||||
yield from 解析Paper(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
|
||||
@ -1,19 +1,115 @@
|
||||
import traceback
|
||||
from toolbox import update_ui
|
||||
|
||||
def request_gpt_model_in_new_thread_with_ui_alive(inputs, inputs_show_user, top_p, temperature, chatbot, history, sys_prompt, refresh_interval=0.2):
|
||||
def input_clipping(inputs, history, max_token_limit):
|
||||
import tiktoken
|
||||
import numpy as np
|
||||
from toolbox import get_conf
|
||||
enc = tiktoken.encoding_for_model(*get_conf('LLM_MODEL'))
|
||||
def get_token_num(txt): return len(enc.encode(txt))
|
||||
|
||||
mode = 'input-and-history'
|
||||
# 当 输入部分的token占比 小于 全文的一半时，只裁剪历史
|
||||
input_token_num = get_token_num(inputs)
|
||||
if input_token_num < max_token_limit//2:
|
||||
mode = 'only-history'
|
||||
max_token_limit = max_token_limit - input_token_num
|
||||
|
||||
everything = [inputs] if mode == 'input-and-history' else ['']
|
||||
everything.extend(history)
|
||||
n_token = get_token_num('\n'.join(everything))
|
||||
everything_token = [get_token_num(e) for e in everything]
|
||||
delta = max(everything_token) // 16 # 截断时的颗粒度
|
||||
|
||||
while n_token > max_token_limit:
|
||||
where = np.argmax(everything_token)
|
||||
encoded = enc.encode(everything[where])
|
||||
clipped_encoded = encoded[:len(encoded)-delta]
|
||||
everything[where] = enc.decode(clipped_encoded)[:-1] # -1 to remove the may-be illegal char
|
||||
everything_token[where] = get_token_num(everything[where])
|
||||
n_token = get_token_num('\n'.join(everything))
|
||||
|
||||
if mode == 'input-and-history':
|
||||
inputs = everything[0]
|
||||
else:
|
||||
pass
|
||||
history = everything[1:]
|
||||
return inputs, history
|
||||
|
||||
def request_gpt_model_in_new_thread_with_ui_alive(
|
||||
inputs, inputs_show_user, top_p, temperature,
|
||||
chatbot, history, sys_prompt, refresh_interval=0.2,
|
||||
handle_token_exceed=True,
|
||||
retry_times_at_unknown_error=2,
|
||||
):
|
||||
"""
|
||||
Request GPT model，请求GPT模型同时维持用户界面活跃。
|
||||
|
||||
输入参数 Args （以_array结尾的输入变量都是列表，列表长度为子任务的数量，执行时，会把列表拆解，放到每个子线程中分别执行）:
|
||||
inputs (string): List of inputs （输入）
|
||||
inputs_show_user (string): List of inputs to show user（展现在报告中的输入，借助此参数，在汇总报告中隐藏啰嗦的真实输入，增强报告的可读性）
|
||||
top_p (float): Top p value for sampling from model distribution （GPT参数，浮点数）
|
||||
temperature (float): Temperature value for sampling from model distribution（GPT参数，浮点数）
|
||||
chatbot: chatbot inputs and outputs （用户界面对话窗口句柄，用于数据流可视化）
|
||||
history (list): List of chat history （历史，对话历史列表）
|
||||
sys_prompt (string): List of system prompts （系统输入，列表，用于输入给GPT的前提提示，比如你是翻译官怎样怎样）
|
||||
refresh_interval (float, optional): Refresh interval for UI (default: 0.2) （刷新时间间隔频率，建议低于1，不可高于3，仅仅服务于视觉效果）
|
||||
handle_token_exceed：是否自动处理token溢出的情况，如果选择自动处理，则会在溢出时暴力截断，默认开启
|
||||
retry_times_at_unknown_error：失败时的重试次数
|
||||
|
||||
输出 Returns:
|
||||
future: 输出，GPT返回的结果
|
||||
"""
|
||||
import time
|
||||
from concurrent.futures import ThreadPoolExecutor
|
||||
from request_llm.bridge_chatgpt import predict_no_ui_long_connection
|
||||
# 用户反馈
|
||||
chatbot.append([inputs_show_user, ""])
|
||||
msg = '正常'
|
||||
yield chatbot, [], msg
|
||||
yield from update_ui(chatbot=chatbot, history=[])
|
||||
executor = ThreadPoolExecutor(max_workers=16)
|
||||
mutable = ["", time.time()]
|
||||
future = executor.submit(lambda:
|
||||
predict_no_ui_long_connection(
|
||||
inputs=inputs, top_p=top_p, temperature=temperature, history=history, sys_prompt=sys_prompt, observe_window=mutable)
|
||||
)
|
||||
def _req_gpt(inputs, history, sys_prompt):
|
||||
retry_op = retry_times_at_unknown_error
|
||||
exceeded_cnt = 0
|
||||
while True:
|
||||
try:
|
||||
# 【第一种情况】：顺利完成
|
||||
result = predict_no_ui_long_connection(
|
||||
inputs=inputs, top_p=top_p, temperature=temperature,
|
||||
history=history, sys_prompt=sys_prompt, observe_window=mutable)
|
||||
return result
|
||||
except ConnectionAbortedError as token_exceeded_error:
|
||||
# 【第二种情况】：Token溢出，
|
||||
if handle_token_exceed:
|
||||
exceeded_cnt += 1
|
||||
# 【选择处理】 尝试计算比例，尽可能多地保留文本
|
||||
from toolbox import get_reduce_token_percent
|
||||
p_ratio, n_exceed = get_reduce_token_percent(str(token_exceeded_error))
|
||||
MAX_TOKEN = 4096
|
||||
EXCEED_ALLO = 512 + 512 * exceeded_cnt
|
||||
inputs, history = input_clipping(inputs, history, max_token_limit=MAX_TOKEN-EXCEED_ALLO)
|
||||
mutable[0] += f'[Local Message] 警告，文本过长将进行截断，Token溢出数：{n_exceed}。\n\n'
|
||||
continue # 返回重试
|
||||
else:
|
||||
# 【选择放弃】
|
||||
tb_str = '```\n' + traceback.format_exc() + '```'
|
||||
mutable[0] += f"[Local Message] 警告，在执行过程中遭遇问题, Traceback：\n\n{tb_str}\n\n"
|
||||
return mutable[0] # 放弃
|
||||
except:
|
||||
# 【第三种情况】：其他错误
|
||||
tb_str = '```\n' + traceback.format_exc() + '```'
|
||||
mutable[0] += f"[Local Message] 警告，在执行过程中遭遇问题, Traceback：\n\n{tb_str}\n\n"
|
||||
if retry_op > 0:
|
||||
retry_op -= 1
|
||||
mutable[0] += f"[Local Message] 重试中 {retry_times_at_unknown_error-retry_op}/{retry_times_at_unknown_error}：\n\n"
|
||||
time.sleep(5)
|
||||
continue # 返回重试
|
||||
else:
|
||||
time.sleep(5)
|
||||
return mutable[0] # 放弃
|
||||
|
||||
future = executor.submit(_req_gpt, inputs, history, sys_prompt)
|
||||
while True:
|
||||
# yield一次以刷新前端页面
|
||||
time.sleep(refresh_interval)
|
||||
@ -27,8 +123,42 @@ def request_gpt_model_in_new_thread_with_ui_alive(inputs, inputs_show_user, top_
|
||||
return future.result()
|
||||
|
||||
|
||||
def request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(inputs_array, inputs_show_user_array, top_p, temperature, chatbot, history_array, sys_prompt_array, refresh_interval=0.2, max_workers=10, scroller_max_len=30):
|
||||
import time
|
||||
def request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
|
||||
inputs_array, inputs_show_user_array, top_p, temperature,
|
||||
chatbot, history_array, sys_prompt_array,
|
||||
refresh_interval=0.2, max_workers=10, scroller_max_len=30,
|
||||
handle_token_exceed=True, show_user_at_complete=False,
|
||||
retry_times_at_unknown_error=2,
|
||||
):
|
||||
"""
|
||||
Request GPT model using multiple threads with UI and high efficiency
|
||||
请求GPT模型的[多线程]版。
|
||||
具备以下功能：
|
||||
实时在UI上反馈远程数据流
|
||||
使用线程池，可调节线程池的大小避免openai的流量限制错误
|
||||
处理中途中止的情况
|
||||
网络等出问题时，会把traceback和已经接收的数据转入输出
|
||||
|
||||
输入参数 Args （以_array结尾的输入变量都是列表，列表长度为子任务的数量，执行时，会把列表拆解，放到每个子线程中分别执行）:
|
||||
inputs_array (list): List of inputs （每个子任务的输入）
|
||||
inputs_show_user_array (list): List of inputs to show user（每个子任务展现在报告中的输入，借助此参数，在汇总报告中隐藏啰嗦的真实输入，增强报告的可读性）
|
||||
top_p (float): Top p value for sampling from model distribution （GPT参数，浮点数）
|
||||
temperature (float): Temperature value for sampling from model distribution（GPT参数，浮点数）
|
||||
chatbot: chatbot （用户界面对话窗口句柄，用于数据流可视化）
|
||||
history_array (list): List of chat history （历史对话输入，双层列表，第一层列表是子任务分解，第二层列表是对话历史）
|
||||
sys_prompt_array (list): List of system prompts （系统输入，列表，用于输入给GPT的前提提示，比如你是翻译官怎样怎样）
|
||||
refresh_interval (float, optional): Refresh interval for UI (default: 0.2) （刷新时间间隔频率，建议低于1，不可高于3，仅仅服务于视觉效果）
|
||||
max_workers (int, optional): Maximum number of threads (default: 10) （最大线程数，如果子任务非常多，需要用此选项防止高频地请求openai导致错误）
|
||||
scroller_max_len (int, optional): Maximum length for scroller (default: 30)（数据流的显示最后收到的多少个字符，仅仅服务于视觉效果）
|
||||
handle_token_exceed (bool, optional): （是否在输入过长时，自动缩减文本）
|
||||
handle_token_exceed：是否自动处理token溢出的情况，如果选择自动处理，则会在溢出时暴力截断，默认开启
|
||||
show_user_at_complete (bool, optional): (在结束时，把完整输入-输出结果显示在聊天框)
|
||||
retry_times_at_unknown_error：子任务失败时的重试次数
|
||||
|
||||
输出 Returns:
|
||||
list: List of GPT model responses （每个子任务的输出汇总，如果某个子任务出错，response中会携带traceback报错信息，方便调试和定位问题。）
|
||||
"""
|
||||
import time, random
|
||||
from concurrent.futures import ThreadPoolExecutor
|
||||
from request_llm.bridge_chatgpt import predict_no_ui_long_connection
|
||||
assert len(inputs_array) == len(history_array)
|
||||
@ -40,20 +170,61 @@ def request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(inp
|
||||
msg = '正常'
|
||||
yield chatbot, [], msg
|
||||
# 异步原子
|
||||
mutable = [["", time.time()] for _ in range(n_frag)]
|
||||
mutable = [["", time.time(), "等待中"] for _ in range(n_frag)]
|
||||
|
||||
def _req_gpt(index, inputs, history, sys_prompt):
|
||||
try:
|
||||
gpt_say = predict_no_ui_long_connection(
|
||||
inputs=inputs, top_p=top_p, temperature=temperature, history=history, sys_prompt=sys_prompt, observe_window=mutable[index]
|
||||
)
|
||||
except:
|
||||
# 收拾残局
|
||||
tb_str = '```\n' + traceback.format_exc() + '```'
|
||||
gpt_say = f"[Local Message] 线程{index}在执行过程中遭遇问题, Traceback：\n\n{tb_str}\n\n"
|
||||
if len(mutable[index][0]) > 0:
|
||||
gpt_say += "此线程失败前收到的回答：" + mutable[index][0]
|
||||
return gpt_say
|
||||
gpt_say = ""
|
||||
retry_op = retry_times_at_unknown_error
|
||||
exceeded_cnt = 0
|
||||
mutable[index][2] = "执行中"
|
||||
while True:
|
||||
try:
|
||||
# 【第一种情况】：顺利完成
|
||||
# time.sleep(10); raise RuntimeError("测试")
|
||||
gpt_say = predict_no_ui_long_connection(
|
||||
inputs=inputs, top_p=top_p, temperature=temperature, history=history,
|
||||
sys_prompt=sys_prompt, observe_window=mutable[index], console_slience=True
|
||||
)
|
||||
mutable[index][2] = "已成功"
|
||||
return gpt_say
|
||||
except ConnectionAbortedError as token_exceeded_error:
|
||||
# 【第二种情况】：Token溢出，
|
||||
if handle_token_exceed:
|
||||
exceeded_cnt += 1
|
||||
# 【选择处理】 尝试计算比例，尽可能多地保留文本
|
||||
from toolbox import get_reduce_token_percent
|
||||
p_ratio, n_exceed = get_reduce_token_percent(str(token_exceeded_error))
|
||||
MAX_TOKEN = 4096
|
||||
EXCEED_ALLO = 512 + 512 * exceeded_cnt
|
||||
inputs, history = input_clipping(inputs, history, max_token_limit=MAX_TOKEN-EXCEED_ALLO)
|
||||
gpt_say += f'[Local Message] 警告，文本过长将进行截断，Token溢出数：{n_exceed}。\n\n'
|
||||
mutable[index][2] = f"截断重试"
|
||||
continue # 返回重试
|
||||
else:
|
||||
# 【选择放弃】
|
||||
tb_str = '```\n' + traceback.format_exc() + '```'
|
||||
gpt_say += f"[Local Message] 警告，线程{index}在执行过程中遭遇问题, Traceback：\n\n{tb_str}\n\n"
|
||||
if len(mutable[index][0]) > 0: gpt_say += "此线程失败前收到的回答：\n\n" + mutable[index][0]
|
||||
mutable[index][2] = "输入过长已放弃"
|
||||
return gpt_say # 放弃
|
||||
except:
|
||||
# 【第三种情况】：其他错误
|
||||
tb_str = '```\n' + traceback.format_exc() + '```'
|
||||
gpt_say += f"[Local Message] 警告，线程{index}在执行过程中遭遇问题, Traceback：\n\n{tb_str}\n\n"
|
||||
if len(mutable[index][0]) > 0: gpt_say += "此线程失败前收到的回答：\n\n" + mutable[index][0]
|
||||
if retry_op > 0:
|
||||
retry_op -= 1
|
||||
wait = random.randint(5, 20)
|
||||
for i in range(wait):# 也许等待十几秒后，情况会好转
|
||||
mutable[index][2] = f"等待重试 {wait-i}"; time.sleep(1)
|
||||
mutable[index][2] = f"重试中 {retry_times_at_unknown_error-retry_op}/{retry_times_at_unknown_error}"
|
||||
continue # 返回重试
|
||||
else:
|
||||
mutable[index][2] = "已失败"
|
||||
wait = 5
|
||||
time.sleep(5)
|
||||
return gpt_say # 放弃
|
||||
|
||||
# 异步任务开始
|
||||
futures = [executor.submit(_req_gpt, index, inputs, history, sys_prompt) for index, inputs, history, sys_prompt in zip(
|
||||
range(len(inputs_array)), inputs_array, history_array, sys_prompt_array)]
|
||||
@ -68,6 +239,7 @@ def request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(inp
|
||||
break
|
||||
# 更好的UI视觉效果
|
||||
observe_win = []
|
||||
# print([mutable[thread_index][2] for thread_index, _ in enumerate(worker_done)])
|
||||
# 每个线程都要“喂狗”（看门狗）
|
||||
for thread_index, _ in enumerate(worker_done):
|
||||
mutable[thread_index][1] = time.time()
|
||||
@ -77,10 +249,10 @@ def request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(inp
|
||||
replace('\n', '').replace('```', '...').replace(
|
||||
' ', '.').replace('<br/>', '.....').replace('$', '.')+"`... ]"
|
||||
observe_win.append(print_something_really_funny)
|
||||
stat_str = ''.join([f'执行中: {obs}\n\n' if not done else '已完成\n\n' for done, obs in zip(
|
||||
worker_done, observe_win)])
|
||||
chatbot[-1] = [chatbot[-1][0],
|
||||
f'多线程操作已经开始，完成情况: \n\n{stat_str}' + ''.join(['.']*(cnt % 10+1))]
|
||||
stat_str = ''.join([f'`{mutable[thread_index][2]}`: {obs}\n\n'
|
||||
if not done else f'`{mutable[thread_index][2]}`\n\n'
|
||||
for thread_index, done, obs in zip(range(len(worker_done)), worker_done, observe_win)])
|
||||
chatbot[-1] = [chatbot[-1][0], f'多线程操作已经开始，完成情况: \n\n{stat_str}' + ''.join(['.']*(cnt % 10+1))]
|
||||
msg = "正常"
|
||||
yield chatbot, [], msg
|
||||
# 异步任务结束
|
||||
@ -88,9 +260,38 @@ def request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(inp
|
||||
for inputs_show_user, f in zip(inputs_show_user_array, futures):
|
||||
gpt_res = f.result()
|
||||
gpt_response_collection.extend([inputs_show_user, gpt_res])
|
||||
|
||||
if show_user_at_complete:
|
||||
for inputs_show_user, f in zip(inputs_show_user_array, futures):
|
||||
gpt_res = f.result()
|
||||
chatbot.append([inputs_show_user, gpt_res])
|
||||
yield chatbot, [], msg
|
||||
time.sleep(1)
|
||||
return gpt_response_collection
|
||||
|
||||
|
||||
def WithRetry(f):
|
||||
"""
|
||||
装饰器函数，用于自动重试。
|
||||
"""
|
||||
def decorated(retry, res_when_fail, *args, **kwargs):
|
||||
assert retry >= 0
|
||||
while True:
|
||||
try:
|
||||
res = yield from f(*args, **kwargs)
|
||||
return res
|
||||
except:
|
||||
retry -= 1
|
||||
if retry<0:
|
||||
print("达到最大重试次数")
|
||||
break
|
||||
else:
|
||||
print("重试中……")
|
||||
continue
|
||||
return res_when_fail
|
||||
return decorated
|
||||
|
||||
|
||||
def breakdown_txt_to_satisfy_token_limit(txt, get_token_fn, limit):
|
||||
def cut(txt_tocut, must_break_at_empty_line): # 递归
|
||||
if get_token_fn(txt_tocut) <= limit:
|
||||
|
||||
@ -58,11 +58,10 @@ def 全项目切换英文(txt, top_p, temperature, chatbot, history, sys_prompt,
|
||||
|
||||
# 第5步：Token限制下的截断与处理
|
||||
MAX_TOKEN = 3000
|
||||
from transformers import GPT2TokenizerFast
|
||||
print('加载tokenizer中')
|
||||
tokenizer = GPT2TokenizerFast.from_pretrained("gpt2")
|
||||
get_token_fn = lambda txt: len(tokenizer(txt)["input_ids"])
|
||||
print('加载tokenizer结束')
|
||||
import tiktoken
|
||||
from toolbox import get_conf
|
||||
enc = tiktoken.encoding_for_model(*get_conf('LLM_MODEL'))
|
||||
def get_token_fn(txt): return len(enc.encode(txt))
|
||||
|
||||
|
||||
# 第6步：任务函数
|
||||
|
||||
@ -148,7 +148,8 @@ def 解析PDF(file_manifest, project_folder, top_p, temperature, chatbot, histor
|
||||
file_content, page_one = read_and_clean_pdf_text(fp)
|
||||
# 递归地切割PDF文件
|
||||
from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
|
||||
enc = tiktoken.get_encoding("gpt2")
|
||||
from toolbox import get_conf
|
||||
enc = tiktoken.encoding_for_model(*get_conf('LLM_MODEL'))
|
||||
def get_token_num(txt): return len(enc.encode(txt))
|
||||
# 分解文本
|
||||
paper_fragments = breakdown_txt_to_satisfy_token_limit_for_pdf(
|
||||
|
||||
@ -2,92 +2,96 @@ from request_llm.bridge_chatgpt import predict_no_ui
|
||||
from toolbox import CatchException, report_execption, write_results_to_file, predict_no_ui_but_counting_down
|
||||
fast_debug = False
|
||||
|
||||
def 解析源代码(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt):
|
||||
import time, glob, os
|
||||
print('begin analysis on:', file_manifest)
|
||||
|
||||
def 解析源代码新(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt):
|
||||
import os, copy
|
||||
from .crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
|
||||
from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive, WithRetry
|
||||
msg = '正常'
|
||||
inputs_array = []
|
||||
inputs_show_user_array = []
|
||||
history_array = []
|
||||
sys_prompt_array = []
|
||||
report_part_1 = []
|
||||
|
||||
############################## <第一步，逐个文件分析，多线程> ##################################
|
||||
for index, fp in enumerate(file_manifest):
|
||||
with open(fp, 'r', encoding='utf-8') as f:
|
||||
file_content = f.read()
|
||||
|
||||
prefix = "接下来请你逐文件分析下面的工程" if index==0 else ""
|
||||
i_say = prefix + f'请对下面的程序文件做一个概述文件名是{os.path.relpath(fp, project_folder)}，文件代码是 ```{file_content}```'
|
||||
i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的程序文件做一个概述: {os.path.abspath(fp)}'
|
||||
chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
|
||||
yield chatbot, history, '正常'
|
||||
# 装载请求内容
|
||||
inputs_array.append(i_say)
|
||||
inputs_show_user_array.append(i_say_show_user)
|
||||
history_array.append([])
|
||||
sys_prompt_array.append("你是一个程序架构分析师，正在分析一个源代码项目。你的回答必须简单明了。")
|
||||
|
||||
if not fast_debug:
|
||||
msg = '正常'
|
||||
gpt_response_collection = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
|
||||
inputs_array = inputs_array,
|
||||
inputs_show_user_array = inputs_show_user_array,
|
||||
history_array = history_array,
|
||||
sys_prompt_array = sys_prompt_array,
|
||||
top_p = top_p,
|
||||
temperature = temperature,
|
||||
chatbot = chatbot,
|
||||
show_user_at_complete = True
|
||||
)
|
||||
|
||||
# ** 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
|
||||
report_part_1 = copy.deepcopy(gpt_response_collection)
|
||||
history_to_return = report_part_1
|
||||
res = write_results_to_file(report_part_1)
|
||||
chatbot.append(("完成？", "逐个文件分析已完成。" + res + "\n\n正在开始汇总。"))
|
||||
yield chatbot, history_to_return, msg
|
||||
|
||||
chatbot[-1] = (i_say_show_user, gpt_say)
|
||||
history.append(i_say_show_user); history.append(gpt_say)
|
||||
yield chatbot, history, msg
|
||||
if not fast_debug: time.sleep(2)
|
||||
|
||||
all_file = ', '.join([os.path.relpath(fp, project_folder) for index, fp in enumerate(file_manifest)])
|
||||
i_say = f'根据以上你自己的分析，对程序的整体功能和构架做出概括。然后用一张markdown表格整理每个文件的功能（包括{all_file}）。'
|
||||
chatbot.append((i_say, "[Local Message] waiting gpt response."))
|
||||
yield chatbot, history, '正常'
|
||||
|
||||
if not fast_debug:
|
||||
msg = '正常'
|
||||
# ** gpt request **
|
||||
gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say, chatbot, top_p, temperature, history=history) # 带超时倒计时
|
||||
|
||||
chatbot[-1] = (i_say, gpt_say)
|
||||
history.append(i_say); history.append(gpt_say)
|
||||
yield chatbot, history, msg
|
||||
res = write_results_to_file(history)
|
||||
chatbot.append(("完成了吗？", res))
|
||||
yield chatbot, history, msg
|
||||
############################## <第二步，综合，单线程，分组+迭代处理> ##################################
|
||||
batchsize = 16 # 10个文件为一组
|
||||
report_part_2 = []
|
||||
previous_iteration_files = []
|
||||
while True:
|
||||
if len(file_manifest) == 0: break
|
||||
this_iteration_file_manifest = file_manifest[:batchsize]
|
||||
this_iteration_gpt_response_collection = gpt_response_collection[:batchsize*2]
|
||||
file_rel_path = [os.path.relpath(fp, project_folder) for index, fp in enumerate(this_iteration_file_manifest)]
|
||||
# 把“请对下面的程序文件做一个概述” 替换成 精简的 "文件名：{all_file[index]}"
|
||||
for index, content in enumerate(this_iteration_gpt_response_collection):
|
||||
if index%2==0: this_iteration_gpt_response_collection[index] = f"文件名：{file_rel_path[index//2]}"
|
||||
previous_iteration_files.extend([os.path.relpath(fp, project_folder) for index, fp in enumerate(this_iteration_file_manifest)])
|
||||
previous_iteration_files_string = ', '.join(previous_iteration_files)
|
||||
current_iteration_focus = ', '.join([os.path.relpath(fp, project_folder) for index, fp in enumerate(this_iteration_file_manifest)])
|
||||
i_say = f'根据以上分析，对程序的整体功能和构架重新做出概括。然后用一张markdown表格整理每个文件的功能（包括{previous_iteration_files_string}）。'
|
||||
inputs_show_user = f'根据以上分析，对程序的整体功能和构架重新做出概括，由于输入长度限制，可能需要分组处理，本组文件为 {current_iteration_focus} + 已经汇总的文件组。'
|
||||
this_iteration_history = copy.deepcopy(this_iteration_gpt_response_collection)
|
||||
this_iteration_history.extend(report_part_2)
|
||||
result = yield from request_gpt_model_in_new_thread_with_ui_alive(
|
||||
inputs=i_say, inputs_show_user=inputs_show_user, top_p=top_p, temperature=temperature, chatbot=chatbot,
|
||||
history=this_iteration_history, # 迭代之前的分析
|
||||
sys_prompt="你是一个程序架构分析师，正在分析一个源代码项目。")
|
||||
report_part_2.extend([i_say, result])
|
||||
|
||||
file_manifest = file_manifest[batchsize:]
|
||||
gpt_response_collection = gpt_response_collection[batchsize*2:]
|
||||
|
||||
############################## <END> ##################################
|
||||
history_to_return.extend(report_part_2)
|
||||
res = write_results_to_file(history_to_return)
|
||||
chatbot.append(("完成了吗？", res))
|
||||
yield chatbot, history_to_return, msg
|
||||
|
||||
|
||||
@CatchException
|
||||
def 解析项目本身(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
|
||||
history = [] # 清空历史，以免输入溢出
|
||||
import time, glob, os
|
||||
import glob
|
||||
file_manifest = [f for f in glob.glob('./*.py') if ('test_project' not in f) and ('gpt_log' not in f)] + \
|
||||
[f for f in glob.glob('./crazy_functions/*.py') if ('test_project' not in f) and ('gpt_log' not in f)]
|
||||
for index, fp in enumerate(file_manifest):
|
||||
# if 'test_project' in fp: continue
|
||||
with open(fp, 'r', encoding='utf-8') as f:
|
||||
file_content = f.read()
|
||||
|
||||
prefix = "接下来请你分析自己的程序构成，别紧张，" if index==0 else ""
|
||||
i_say = prefix + f'请对下面的程序文件做一个概述文件名是{fp}，文件代码是 ```{file_content}```'
|
||||
i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的程序文件做一个概述: {os.path.abspath(fp)}'
|
||||
chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
|
||||
yield chatbot, history, '正常'
|
||||
|
||||
if not fast_debug:
|
||||
# ** gpt request **
|
||||
# gpt_say = predict_no_ui(inputs=i_say, top_p=top_p, temperature=temperature)
|
||||
gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temperature, history=[], long_connection=True) # 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, '正常'
|
||||
time.sleep(2)
|
||||
|
||||
i_say = f'根据以上你自己的分析，对程序的整体功能和构架做出概括。然后用一张markdown表格整理每个文件的功能（包括{file_manifest}）。'
|
||||
chatbot.append((i_say, "[Local Message] waiting gpt response."))
|
||||
yield chatbot, history, '正常'
|
||||
|
||||
if not fast_debug:
|
||||
# ** gpt request **
|
||||
# gpt_say = predict_no_ui(inputs=i_say, top_p=top_p, temperature=temperature, history=history)
|
||||
gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say, chatbot, top_p, temperature, history=history, long_connection=True) # 带超时倒计时
|
||||
|
||||
chatbot[-1] = (i_say, gpt_say)
|
||||
history.append(i_say); history.append(gpt_say)
|
||||
yield chatbot, history, '正常'
|
||||
res = write_results_to_file(history)
|
||||
chatbot.append(("完成了吗？", res))
|
||||
[f for f in glob.glob('./crazy_functions/*.py') if ('test_project' not in f) and ('gpt_log' not in f)]+ \
|
||||
[f for f in glob.glob('./request_llm/*.py') if ('test_project' not in f) and ('gpt_log' not in f)]
|
||||
project_folder = './'
|
||||
if len(file_manifest) == 0:
|
||||
report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何python文件: {txt}")
|
||||
yield chatbot, history, '正常'
|
||||
return
|
||||
yield from 解析源代码新(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
|
||||
|
||||
@CatchException
|
||||
def 解析一个Python项目(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
|
||||
@ -105,7 +109,7 @@ def 解析一个Python项目(txt, top_p, temperature, chatbot, history, systemPr
|
||||
report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何python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)
|
||||
yield from 解析源代码新(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
|
||||
|
||||
|
||||
@CatchException
|
||||
@ -126,7 +130,7 @@ def 解析一个C项目的头文件(txt, top_p, temperature, chatbot, history, s
|
||||
report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.h头文件: {txt}")
|
||||
yield chatbot, history, '正常'
|
||||
return
|
||||
yield from 解析源代码(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
|
||||
yield from 解析源代码新(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
|
||||
|
||||
@CatchException
|
||||
def 解析一个C项目(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
|
||||
@ -147,7 +151,7 @@ def 解析一个C项目(txt, top_p, temperature, chatbot, history, systemPromptT
|
||||
report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.h头文件: {txt}")
|
||||
yield chatbot, history, '正常'
|
||||
return
|
||||
yield from 解析源代码(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
|
||||
yield from 解析源代码新(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
|
||||
|
||||
|
||||
@CatchException
|
||||
@ -169,7 +173,7 @@ def 解析一个Java项目(txt, top_p, temperature, chatbot, history, systemProm
|
||||
report_execption(chatbot, history, a=f"解析项目: {txt}", b=f"找不到任何java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)
|
||||
yield from 解析源代码新(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
|
||||
|
||||
|
||||
@CatchException
|
||||
@ -192,7 +196,7 @@ def 解析一个Rect项目(txt, top_p, temperature, chatbot, history, systemProm
|
||||
report_execption(chatbot, history, a=f"解析项目: {txt}", b=f"找不到任何Rect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)
|
||||
yield from 解析源代码新(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
|
||||
|
||||
|
||||
@CatchException
|
||||
@ -211,4 +215,4 @@ def 解析一个Golang项目(txt, top_p, temperature, chatbot, history, systemPr
|
||||
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)
|
||||
yield from 解析源代码新(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
|
||||
|
||||
BIN
objdump.tmp
Normal file
BIN
objdump.tmp
Normal file
Binary file not shown.
@ -72,7 +72,7 @@ def predict_no_ui(inputs, top_p, temperature, history=[], sys_prompt=""):
|
||||
raise ConnectionAbortedError("Json解析不合常规，可能是文本过长" + response.text)
|
||||
|
||||
|
||||
def predict_no_ui_long_connection(inputs, top_p, temperature, history=[], sys_prompt="", observe_window=None):
|
||||
def predict_no_ui_long_connection(inputs, top_p, temperature, history=[], sys_prompt="", observe_window=None, console_slience=False):
|
||||
"""
|
||||
发送至chatGPT，等待回复，一次性完成，不显示中间过程。但内部用stream的方法避免中途网线被掐。
|
||||
inputs：
|
||||
@ -121,7 +121,7 @@ def predict_no_ui_long_connection(inputs, top_p, temperature, history=[], sys_pr
|
||||
if "role" in delta: continue
|
||||
if "content" in delta:
|
||||
result += delta["content"]
|
||||
print(delta["content"], end='')
|
||||
if not console_slience: print(delta["content"], end='')
|
||||
if observe_window is not None:
|
||||
# 观测窗，把已经获取的数据显示出去
|
||||
if len(observe_window) >= 1: observe_window[0] += delta["content"]
|
||||
@ -264,8 +264,7 @@ def generate_payload(inputs, top_p, temperature, history, system_prompt, stream)
|
||||
"presence_penalty": 0,
|
||||
"frequency_penalty": 0,
|
||||
}
|
||||
|
||||
print(f" {LLM_MODEL} : {conversation_cnt} : {inputs}")
|
||||
print(f" {LLM_MODEL} : {conversation_cnt} : {inputs[:100]}")
|
||||
return headers,payload
|
||||
|
||||
|
||||
|
||||
@ -21,6 +21,8 @@ def ArgsGeneralWrapper(f):
|
||||
yield from f(txt_passon, *args, **kwargs)
|
||||
return decorated
|
||||
|
||||
def update_ui(chatbot, history, msg='正常', *args, **kwargs):
|
||||
yield chatbot, history, msg
|
||||
|
||||
def get_reduce_token_percent(text):
|
||||
try:
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user