From 85d85d850a25bb3c380e33ab278fcc1ce7e4b99c Mon Sep 17 00:00:00 2001 From: Your Name Date: Thu, 6 Apr 2023 03:30:02 +0800 Subject: [PATCH] update --- crazy_functions/crazy_utils.py | 11 +- crazy_functions/批量翻译PDF文档_多线程.py | 191 ++++++++++++---------- crazy_functions/高级功能函数模板.py | 2 +- 3 files changed, 119 insertions(+), 85 deletions(-) diff --git a/crazy_functions/crazy_utils.py b/crazy_functions/crazy_utils.py index f1b5e82..49732f3 100644 --- a/crazy_functions/crazy_utils.py +++ b/crazy_functions/crazy_utils.py @@ -37,6 +37,7 @@ def breakdown_txt_to_satisfy_token_limit_for_pdf(txt, get_token_fn, limit): lines = txt_tocut.split('\n') estimated_line_cut = limit / get_token_fn(txt_tocut) * len(lines) estimated_line_cut = int(estimated_line_cut) + cnt = 0 for cnt in reversed(range(estimated_line_cut)): if must_break_at_empty_line: if lines[cnt] != "": continue @@ -45,7 +46,7 @@ def breakdown_txt_to_satisfy_token_limit_for_pdf(txt, get_token_fn, limit): post = "\n".join(lines[cnt:]) if get_token_fn(prev) < limit: break if cnt == 0: - print('what the fuck ?') + # print('what the fuck ? 存在一行极长的文本！') raise RuntimeError("存在一行极长的文本！") # print(len(post)) # 列表递归接龙 @@ -55,4 +56,10 @@ def breakdown_txt_to_satisfy_token_limit_for_pdf(txt, get_token_fn, limit): try: return cut(txt, must_break_at_empty_line=True) except RuntimeError: - return cut(txt, must_break_at_empty_line=False) + try: + return cut(txt, must_break_at_empty_line=False) + except RuntimeError: + # 这个中文的句号是故意的，作为一个标识而存在 + res = cut(txt.replace('.', '。\n'), must_break_at_empty_line=False) + return [r.replace('。\n', '.') for r in res] + diff --git a/crazy_functions/批量翻译PDF文档_多线程.py b/crazy_functions/批量翻译PDF文档_多线程.py index 91ad003..c8f72b8 100644 --- a/crazy_functions/批量翻译PDF文档_多线程.py +++ b/crazy_functions/批量翻译PDF文档_多线程.py @@ -1,7 +1,6 @@ from toolbox import CatchException, report_execption, write_results_to_file, predict_no_ui_but_counting_down import re import unicodedata -fast_debug = False def is_paragraph_break(match): @@ -61,7 +60,6 @@ def clean_text(raw_text): return final_text.strip() - def read_and_clean_pdf_text(fp): import fitz, re import numpy as np @@ -69,19 +67,16 @@ def read_and_clean_pdf_text(fp): with fitz.open(fp) as doc: meta_txt = [] meta_font = [] - for page in doc: + for index, page in enumerate(doc): # file_content += page.get_text() text_areas = page.get_text("dict") # 获取页面上的文本信息 + # 块元提取 for each word segment with in line for each line cross-line words for each block + meta_txt.extend( [ " ".join(["".join( [wtf['text'] for wtf in l['spans'] ]) for l in t['lines'] ]).replace('- ','') for t in text_areas['blocks'] if 'lines' in t]) + meta_font.extend([ np.mean( [ np.mean([wtf['size'] for wtf in l['spans'] ]) for l in t['lines'] ]) for t in text_areas['blocks'] if 'lines' in t]) + if index==0: + page_one_meta = [" ".join(["".join( [wtf['text'] for wtf in l['spans'] ]) for l in t['lines'] ]).replace('- ','') for t in text_areas['blocks'] if 'lines' in t] - # # 行元提取 for each word segment with in line for each line for each block - # meta_txt.extend( [ ["".join( [wtf['text'] for wtf in l['spans'] ]) for l in t['lines'] ] for t in text_areas['blocks'] if 'lines' in t]) - # meta_font.extend([ [ np.mean([wtf['size'] for wtf in l['spans'] ]) for l in t['lines'] ] for t in text_areas['blocks'] if 'lines' in t]) - - # 块元提取 for each word segment with in line for each line for each block - meta_txt.extend( [ " ".join(["".join( [wtf['text'] for wtf in l['spans'] ]) for l in t['lines'] ]) for t in text_areas['blocks'] if 'lines' in t]) - meta_font.extend([ np.mean( [ np.mean([wtf['size'] for wtf in l['spans'] ]) for l in t['lines'] ]) for t in text_areas['blocks'] if 'lines' in t]) - def 把字符太少的块清除为回车(meta_txt): for index, block_txt in enumerate(meta_txt): if len(block_txt) < 100: @@ -123,19 +118,17 @@ def read_and_clean_pdf_text(fp): # 换行 -> 双换行 meta_txt = meta_txt.replace('\n', '\n\n') - # print(meta_txt) - - return meta_txt + return meta_txt, page_one_meta @CatchException -def 批量翻译PDF文档(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT): +def 批量翻译PDF文档(txt, top_p, temperature, chatbot, history, sys_prompt, WEB_PORT): import glob import os # 基本信息：功能、贡献者 chatbot.append([ "函数插件功能？", - "批量总结PDF文档。函数插件贡献者: Binary-Husky, ValeriaWong, Eralien"]) + "批量总结PDF文档。函数插件贡献者: Binary-Husky（二进制哈士奇）"]) yield chatbot, history, '正常' # 尝试导入依赖，如果缺少依赖，则给出安装建议 @@ -174,82 +167,116 @@ def 批量翻译PDF文档(txt, top_p, temperature, chatbot, history, systemPromp return # 开始正式执行任务 - yield from 解析PDF(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt) + yield from 解析PDF(file_manifest, project_folder, top_p, temperature, chatbot, history, sys_prompt) -def 解析PDF(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt): +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): + 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 + 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) + ) + while True: + # yield一次以刷新前端页面 + time.sleep(refresh_interval) + # “喂狗”（看门狗） + mutable[1] = time.time() + if future.done(): break + chatbot[-1] = [chatbot[-1][0], mutable[0]]; msg = "正常" + yield chatbot, [], msg + 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, max_workers=10, scroller_max_len=30): + import time + from concurrent.futures import ThreadPoolExecutor + from request_llm.bridge_chatgpt import predict_no_ui_long_connection + assert len(inputs_array) == len(history_array) + assert len(inputs_array) == len(sys_prompt_array) + executor = ThreadPoolExecutor(max_workers=max_workers) + n_frag = len(inputs_array) + # 异步原子 + mutable = [["", time.time()] for _ in range(n_frag)] + def _req_gpt(index, inputs, history, sys_prompt): + 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] + ) + return gpt_say + # 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)] + cnt = 0 + while True: + # yield一次以刷新前端页面 + time.sleep(refresh_interval); cnt += 1 + worker_done = [h.done() for h in futures] + if all(worker_done): executor.shutdown(); break + # 更好的UI视觉效果 + observe_win = [] + # 每个线程都要“喂狗”（看门狗） + for thread_index, _ in enumerate(worker_done): mutable[thread_index][1] = time.time() + # 在前端打印些好玩的东西 + for thread_index, _ in enumerate(worker_done): + print_something_really_funny = "[ ...`"+mutable[thread_index][0][-scroller_max_len:].\ + replace('\n','').replace('```','...').replace(' ','.').replace('
','.....').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))]; msg = "正常" + yield chatbot, [], msg + # 异步任务结束 + gpt_response_collection = [] + 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]) + return gpt_response_collection + +def 解析PDF(file_manifest, project_folder, top_p, temperature, chatbot, history, sys_prompt): import time import glob import os import fitz import tiktoken - from concurrent.futures import ThreadPoolExecutor - print('begin analysis on:', file_manifest) + TOKEN_LIMIT_PER_FRAGMENT = 1600 + for index, fp in enumerate(file_manifest): - ### 1. 读取PDF文件 - file_content = read_and_clean_pdf_text(fp) - ### 2. 递归地切割PDF文件 + # 读取PDF文件 + 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") - TOKEN_LIMIT_PER_FRAGMENT = 2048 get_token_num = lambda txt: len(enc.encode(txt)) - # 分解 - paper_fragments = breakdown_txt_to_satisfy_token_limit_for_pdf( - txt=file_content, get_token_fn=get_token_num, limit=TOKEN_LIMIT_PER_FRAGMENT) - print([get_token_num(frag) for frag in paper_fragments]) - ### 3. 逐个段落翻译 - ## 3.1. 多线程开始 - from request_llm.bridge_chatgpt import predict_no_ui_long_connection - n_frag = len(paper_fragments) - # 异步原子 - mutable = [["", time.time()] for _ in range(n_frag)] - # 翻译函数 - def translate_(index, fragment, mutable): - i_say = f"以下是你需要翻译的文章段落：{fragment}" - # 请求gpt，需要一段时间 - gpt_say = predict_no_ui_long_connection( - inputs=i_say, top_p=top_p, temperature=temperature, history=[], # ["请翻译：" if len(previous_result)!=0 else "", previous_result], - sys_prompt="请你作为一个学术翻译，负责将给定的文章段落翻译成中文，要求语言简洁、精准、凝练。你只需要给出翻译后的文本，不能重复原文。", - observe_window=mutable[index]) - return gpt_say - ### 4. 异步任务开始 - executor = ThreadPoolExecutor(max_workers=16) - # Submit tasks to the pool - futures = [executor.submit(translate_, index, frag, mutable) for index, frag in enumerate(paper_fragments)] - - ### 5. UI主线程，在任务期间提供实时的前端显示 - cnt = 0 - while True: - cnt += 1 - time.sleep(1) - worker_done = [h.done() for h in futures] - if all(worker_done): - executor.shutdown(); break - # 更好的UI视觉效果 - observe_win = [] - # 每个线程都要喂狗（看门狗） - for thread_index, _ in enumerate(worker_done): - mutable[thread_index][1] = time.time() - # 在前端打印些好玩的东西 - for thread_index, _ in enumerate(worker_done): - print_something_really_funny = "[ ...`"+mutable[thread_index][0][-30:].replace('\n','').replace('```','...').replace(' ','.').replace('
','.....').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))]; msg = "正常" - yield chatbot, history, msg + # 分解文本 + paper_fragments = breakdown_txt_to_satisfy_token_limit_for_pdf( + txt=file_content, get_token_fn=get_token_num, limit=TOKEN_LIMIT_PER_FRAGMENT) + page_one_fragments = breakdown_txt_to_satisfy_token_limit_for_pdf( + txt=str(page_one), get_token_fn=get_token_num, limit=TOKEN_LIMIT_PER_FRAGMENT//4) + # 为了更好的效果，我们剥离Introduction之后的部分 + paper_meta = page_one_fragments[0].split('introduction')[0].split('Introduction')[0].split('INTRODUCTION')[0] + # 单线，获取文章meta信息 + paper_meta_info = yield from request_gpt_model_in_new_thread_with_ui_alive( + inputs=f"以下是一篇学术论文的基础信息，请从中提取出“标题”、“收录会议或期刊”、“作者”、“摘要”、“编号”、“作者邮箱”这六个部分。请用markdown格式输出，最后用中文翻译摘要部分。请提取：{paper_meta}", + inputs_show_user=f"请从{fp}中提取出“标题”、“收录会议或期刊”等基本信息。", + top_p=top_p, temperature=temperature, + chatbot=chatbot, history=[], + sys_prompt="Your job is to collect information from materials。", + ) + # 多线，翻译 + gpt_response_collection = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency( + inputs_array = [f"以下是你需要翻译的文章段落：\n{frag}" for frag in paper_fragments], + inputs_show_user_array = [f"" for _ in paper_fragments], + top_p=top_p, temperature=temperature, + chatbot=chatbot, + history_array=[[paper_meta] for _ in paper_fragments], + sys_prompt_array=["请你作为一个学术翻译，把整个段落翻译成中文，要求语言简洁，禁止重复输出原文。" for _ in paper_fragments], + max_workers=16 # OpenAI所允许的最大并行过载 + ) - # Wait for tasks to complete - results = [future.result() for future in futures] + final = ["", paper_meta_info + '\n\n---\n\n---\n\n---\n\n'].extend(gpt_response_collection) + res = write_results_to_file(final) + chatbot.append((f"{fp}完成了吗？", res)); msg = "完成" + yield chatbot, history, msg - print(results) - # full_result += gpt_say - - # history.extend([fp, full_result]) - - res = write_results_to_file(history) - chatbot.append(("完成了吗？", res)); msg = "完成" - yield chatbot, history, msg - - -# if __name__ == '__main__': -# pro() diff --git a/crazy_functions/高级功能函数模板.py b/crazy_functions/高级功能函数模板.py index 69fd379..95e8240 100644 --- a/crazy_functions/高级功能函数模板.py +++ b/crazy_functions/高级功能函数模板.py @@ -5,7 +5,7 @@ import datetime @CatchException def 高阶功能模板函数(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT): history = [] # 清空历史，以免输入溢出 - chatbot.append(("这是什么功能？", "[Local Message] 请注意，您正在调用一个[函数插件]的模板，该函数面向希望实现更多有趣功能的开发者，它可以作为创建新功能函数的模板。为了做到简单易读，该函数只有25行代码，所以不会实时反馈文字流或心跳，请耐心等待程序输出完成。此外我们也提供可同步处理大量文件的多线程Demo供您参考。您若希望分享新的功能模组，请不吝PR！")) + chatbot.append(("这是什么功能？", "[Local Message] 请注意，您正在调用一个[函数插件]的模板，该函数面向希望实现更多有趣功能的开发者，它可以作为创建新功能函数的模板(该函数只有25行代码)。此外我们也提供可同步处理大量文件的多线程Demo供您参考。您若希望分享新的功能模组，请不吝PR！")) yield chatbot, history, '正常' # 由于请求gpt需要一段时间，我们先及时地做一次状态显示 for i in range(5):