支持借助GROBID实现PDF高精度翻译
This commit is contained in:
binary-husky
parent
3f4c4ebc29
commit
ed11269aef
28
config.py
28
config.py
@ -70,8 +70,10 @@ MAX_RETRY = 2
|
||||
|
||||
# 模型选择是 (注意: LLM_MODEL是默认选中的模型, 它*必须*被包含在AVAIL_LLM_MODELS列表中 )
|
||||
LLM_MODEL = "gpt-3.5-turbo" # 可选 ↓↓↓
|
||||
AVAIL_LLM_MODELS = ["gpt-3.5-turbo-16k", "gpt-3.5-turbo", "azure-gpt-3.5", "api2d-gpt-3.5-turbo", "gpt-4", "api2d-gpt-4", "chatglm", "moss", "newbing", "stack-claude"]
|
||||
# P.S. 其他可用的模型还包括 ["qianfan", "llama2", "qwen", "gpt-3.5-turbo-0613", "gpt-3.5-turbo-16k-0613", "spark", "chatglm_onnx", "claude-1-100k", "claude-2", "internlm", "jittorllms_rwkv", "jittorllms_pangualpha", "jittorllms_llama"]
|
||||
AVAIL_LLM_MODELS = ["gpt-3.5-turbo-16k", "gpt-3.5-turbo", "azure-gpt-3.5", "api2d-gpt-3.5-turbo",
|
||||
"gpt-4", "api2d-gpt-4", "chatglm", "moss", "newbing", "stack-claude"]
|
||||
# P.S. 其他可用的模型还包括 ["qianfan", "llama2", "qwen", "gpt-3.5-turbo-0613", "gpt-3.5-turbo-16k-0613",
|
||||
# "spark", "chatglm_onnx", "claude-1-100k", "claude-2", "internlm", "jittorllms_pangualpha", "jittorllms_llama"]
|
||||
|
||||
|
||||
# 百度千帆（LLM_MODEL="qianfan"）
|
||||
@ -162,6 +164,14 @@ CUSTOM_API_KEY_PATTERN = ""
|
||||
HUGGINGFACE_ACCESS_TOKEN = "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"
|
||||
|
||||
|
||||
# GROBID服务器地址（填写多个可以均衡负载），用于高质量地读取PDF文档
|
||||
# 获取方法：复制以下空间https://huggingface.co/spaces/qingxu98/grobid，设为public，然后GROBID_URL = "https://(你的hf用户名如qingxu98)-(你的填写的空间名如grobid).hf.space"
|
||||
GROBID_URLS = [
|
||||
"https://qingxu98-grobid.hf.space","https://qingxu98-grobid2.hf.space","https://qingxu98-grobid3.hf.space",
|
||||
"https://shaocongma-grobid.hf.space","https://FBR123-grobid.hf.space",
|
||||
]
|
||||
|
||||
|
||||
|
||||
"""
|
||||
在线大模型配置关联关系示意图
|
||||
@ -205,9 +215,13 @@ HUGGINGFACE_ACCESS_TOKEN = "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"
|
||||
插件在线服务配置依赖关系示意图
|
||||
│
|
||||
├── 语音功能
|
||||
├── ENABLE_AUDIO
|
||||
├── ALIYUN_TOKEN
|
||||
├── ALIYUN_APPKEY
|
||||
├── ALIYUN_ACCESSKEY
|
||||
└── ALIYUN_SECRET
|
||||
│ ├── ENABLE_AUDIO
|
||||
│ ├── ALIYUN_TOKEN
|
||||
│ ├── ALIYUN_APPKEY
|
||||
│ ├── ALIYUN_ACCESSKEY
|
||||
│ └── ALIYUN_SECRET
|
||||
│
|
||||
├── PDF文档精准解析
|
||||
│ └── GROBID_URLS
|
||||
|
||||
"""
|
||||
25
crazy_functions/pdf_fns/parse_pdf.py
Normal file
25
crazy_functions/pdf_fns/parse_pdf.py
Normal file
@ -0,0 +1,25 @@
|
||||
import requests
|
||||
import random
|
||||
from functools import lru_cache
|
||||
class GROBID_OFFLINE_EXCEPTION(Exception): pass
|
||||
|
||||
def get_avail_grobid_url():
|
||||
from toolbox import get_conf
|
||||
GROBID_URLS, = get_conf('GROBID_URLS')
|
||||
if len(GROBID_URLS) == 0: return None
|
||||
try:
|
||||
_grobid_url = random.choice(GROBID_URLS) # 随机负载均衡
|
||||
if _grobid_url.endswith('/'): _grobid_url = _grobid_url.rstrip('/')
|
||||
res = requests.get(_grobid_url+'/api/isalive')
|
||||
if res.text=='true': return _grobid_url
|
||||
else: return None
|
||||
except:
|
||||
return None
|
||||
|
||||
@lru_cache(maxsize=32)
|
||||
def parse_pdf(pdf_path, grobid_url):
|
||||
import scipdf # pip install scipdf_parser
|
||||
if grobid_url.endswith('/'): grobid_url = grobid_url.rstrip('/')
|
||||
article_dict = scipdf.parse_pdf_to_dict(pdf_path, grobid_url=grobid_url)
|
||||
return article_dict
|
||||
|
||||
@ -1,15 +1,19 @@
|
||||
from toolbox import CatchException, report_execption, write_results_to_file
|
||||
from toolbox import update_ui, promote_file_to_downloadzone
|
||||
from toolbox import update_ui, promote_file_to_downloadzone, update_ui_lastest_msg, disable_auto_promotion
|
||||
from toolbox import write_history_to_file, get_log_folder
|
||||
from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive
|
||||
from .crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
|
||||
from .crazy_utils import read_and_clean_pdf_text
|
||||
from .pdf_fns.parse_pdf import parse_pdf, get_avail_grobid_url
|
||||
from colorful import *
|
||||
import glob
|
||||
import os
|
||||
import math
|
||||
|
||||
@CatchException
|
||||
def 批量翻译PDF文档(txt, llm_kwargs, plugin_kwargs, chatbot, history, sys_prompt, web_port):
|
||||
import glob
|
||||
import os
|
||||
def 批量翻译PDF文档(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
|
||||
|
||||
disable_auto_promotion(chatbot)
|
||||
# 基本信息：功能、贡献者
|
||||
chatbot.append([
|
||||
"函数插件功能？",
|
||||
@ -30,20 +34,11 @@ def 批量翻译PDF文档(txt, llm_kwargs, plugin_kwargs, chatbot, history, sys_
|
||||
# 清空历史，以免输入溢出
|
||||
history = []
|
||||
|
||||
from .crazy_utils import get_files_from_everything
|
||||
success, file_manifest, project_folder = get_files_from_everything(txt, type='.pdf')
|
||||
# 检测输入参数，如没有给定输入参数，直接退出
|
||||
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 from update_ui(chatbot=chatbot, history=history) # 刷新界面
|
||||
return
|
||||
|
||||
# 搜索需要处理的文件清单
|
||||
file_manifest = [f for f in glob.glob(
|
||||
f'{project_folder}/**/*.pdf', recursive=True)]
|
||||
if not success:
|
||||
if txt == "": txt = '空空如也的输入栏'
|
||||
|
||||
# 如果没找到任何文件
|
||||
if len(file_manifest) == 0:
|
||||
@ -53,22 +48,130 @@ def 批量翻译PDF文档(txt, llm_kwargs, plugin_kwargs, chatbot, history, sys_
|
||||
return
|
||||
|
||||
# 开始正式执行任务
|
||||
yield from 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, sys_prompt)
|
||||
grobid_url = get_avail_grobid_url()
|
||||
if grobid_url is not None:
|
||||
yield from 解析PDF_基于GROBID(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, grobid_url)
|
||||
else:
|
||||
yield from update_ui_lastest_msg("GROBID服务不可用，请检查config中的GROBID_URL。作为替代，现在将执行效果稍差的旧版代码。", chatbot, history, delay=3)
|
||||
yield from 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt)
|
||||
|
||||
|
||||
def 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, sys_prompt):
|
||||
import os
|
||||
def 解析PDF_基于GROBID(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, grobid_url):
|
||||
import copy
|
||||
import tiktoken
|
||||
TOKEN_LIMIT_PER_FRAGMENT = 1280
|
||||
generated_conclusion_files = []
|
||||
generated_html_files = []
|
||||
DST_LANG = "中文"
|
||||
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)
|
||||
print(article_dict)
|
||||
prompt = "以下是一篇学术论文的基本信息:\n"
|
||||
# title
|
||||
title = article_dict.get('title', '无法获取 title'); prompt += f'title:{title}\n\n'
|
||||
# authors
|
||||
authors = article_dict.get('authors', '无法获取 authors'); prompt += f'authors:{authors}\n\n'
|
||||
# abstract
|
||||
abstract = article_dict.get('abstract', '无法获取 abstract'); prompt += f'abstract:{abstract}\n\n'
|
||||
# command
|
||||
prompt += f"请将题目和摘要翻译为{DST_LANG}。"
|
||||
meta = [f'# Title:\n\n', title, f'# Abstract:\n\n', abstract ]
|
||||
|
||||
# 单线，获取文章meta信息
|
||||
paper_meta_info = yield from request_gpt_model_in_new_thread_with_ui_alive(
|
||||
inputs=prompt,
|
||||
inputs_show_user=prompt,
|
||||
llm_kwargs=llm_kwargs,
|
||||
chatbot=chatbot, history=[],
|
||||
sys_prompt="You are an academic paper reader。",
|
||||
)
|
||||
|
||||
# 多线，翻译
|
||||
inputs_array = []
|
||||
inputs_show_user_array = []
|
||||
|
||||
# get_token_num
|
||||
from request_llm.bridge_all import model_info
|
||||
enc = model_info[llm_kwargs['llm_model']]['tokenizer']
|
||||
def get_token_num(txt): return len(enc.encode(txt, disallowed_special=()))
|
||||
from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
|
||||
|
||||
def break_down(txt):
|
||||
raw_token_num = get_token_num(txt)
|
||||
if raw_token_num <= TOKEN_LIMIT_PER_FRAGMENT:
|
||||
return [txt]
|
||||
else:
|
||||
# raw_token_num > TOKEN_LIMIT_PER_FRAGMENT
|
||||
# find a smooth token limit to achieve even seperation
|
||||
count = int(math.ceil(raw_token_num / TOKEN_LIMIT_PER_FRAGMENT))
|
||||
token_limit_smooth = raw_token_num // count + count
|
||||
return breakdown_txt_to_satisfy_token_limit_for_pdf(txt, get_token_fn=get_token_num, limit=token_limit_smooth)
|
||||
|
||||
for section in article_dict.get('sections'):
|
||||
if len(section['text']) == 0: continue
|
||||
section_frags = break_down(section['text'])
|
||||
for i, fragment in enumerate(section_frags):
|
||||
heading = section['heading']
|
||||
if len(section_frags) > 1: heading += f'Part-{i+1}'
|
||||
inputs_array.append(
|
||||
f"你需要翻译{heading}章节，内容如下: \n\n{fragment}"
|
||||
)
|
||||
inputs_show_user_array.append(
|
||||
f"# {heading}\n\n{fragment}"
|
||||
)
|
||||
|
||||
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,
|
||||
llm_kwargs=llm_kwargs,
|
||||
chatbot=chatbot,
|
||||
history_array=[meta for _ in inputs_array],
|
||||
sys_prompt_array=[
|
||||
"请你作为一个学术翻译，负责把学术论文准确翻译成中文。注意文章中的每一句话都要翻译。" for _ in inputs_array],
|
||||
)
|
||||
res_path = write_history_to_file(meta + ["# Meta Translation" , paper_meta_info] + gpt_response_collection, file_basename=None, file_fullname=None)
|
||||
promote_file_to_downloadzone(res_path, rename_file=os.path.basename(fp)+'.md', chatbot=chatbot)
|
||||
generated_conclusion_files.append(res_path)
|
||||
|
||||
ch = construct_html()
|
||||
orig = ""
|
||||
trans = ""
|
||||
gpt_response_collection_html = copy.deepcopy(gpt_response_collection)
|
||||
for i,k in enumerate(gpt_response_collection_html):
|
||||
if i%2==0:
|
||||
gpt_response_collection_html[i] = inputs_show_user_array[i//2]
|
||||
else:
|
||||
gpt_response_collection_html[i] = gpt_response_collection_html[i]
|
||||
|
||||
final = ["", "", "一、论文概况", "", "Abstract", paper_meta_info, "二、论文翻译", ""]
|
||||
final.extend(gpt_response_collection_html)
|
||||
for i, k in enumerate(final):
|
||||
if i%2==0:
|
||||
orig = k
|
||||
if i%2==1:
|
||||
trans = k
|
||||
ch.add_row(a=orig, b=trans)
|
||||
create_report_file_name = f"{os.path.basename(fp)}.trans.html"
|
||||
html_file = ch.save_file(create_report_file_name)
|
||||
generated_html_files.append(html_file)
|
||||
promote_file_to_downloadzone(html_file, rename_file=os.path.basename(html_file), chatbot=chatbot)
|
||||
|
||||
chatbot.append(("给出输出文件清单", str(generated_conclusion_files + generated_html_files)))
|
||||
yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
|
||||
|
||||
|
||||
def 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt):
|
||||
import copy
|
||||
TOKEN_LIMIT_PER_FRAGMENT = 1280
|
||||
generated_conclusion_files = []
|
||||
generated_html_files = []
|
||||
for index, fp in enumerate(file_manifest):
|
||||
# 读取PDF文件
|
||||
file_content, page_one = read_and_clean_pdf_text(fp)
|
||||
file_content = file_content.encode('utf-8', 'ignore').decode() # avoid reading non-utf8 chars
|
||||
page_one = str(page_one).encode('utf-8', 'ignore').decode() # avoid reading non-utf8 chars
|
||||
|
||||
# 递归地切割PDF文件
|
||||
from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
|
||||
from request_llm.bridge_all import model_info
|
||||
@ -140,8 +243,7 @@ def 解析PDF(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot,
|
||||
trans = k
|
||||
ch.add_row(a=orig, b=trans)
|
||||
create_report_file_name = f"{os.path.basename(fp)}.trans.html"
|
||||
ch.save_file(create_report_file_name)
|
||||
generated_html_files.append(f'./gpt_log/{create_report_file_name}')
|
||||
generated_html_files.append(ch.save_file(create_report_file_name))
|
||||
except:
|
||||
from toolbox import trimmed_format_exc
|
||||
print('writing html result failed:', trimmed_format_exc())
|
||||
@ -202,6 +304,6 @@ class construct_html():
|
||||
|
||||
|
||||
def save_file(self, file_name):
|
||||
with open(f'./gpt_log/{file_name}', 'w', encoding='utf8') as f:
|
||||
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)
|
||||
|
||||
@ -18,5 +18,6 @@ openai
|
||||
numpy
|
||||
arxiv
|
||||
rich
|
||||
websocket-client
|
||||
pypdf2==2.12.1
|
||||
websocket-client
|
||||
scipdf_parser==0.3
|
||||
|
||||
@ -9,9 +9,9 @@ validate_path() # 返回项目根路径
|
||||
from tests.test_utils import plugin_test
|
||||
|
||||
if __name__ == "__main__":
|
||||
plugin_test(plugin='crazy_functions.命令行助手->命令行助手', main_input='查看当前的docker容器列表')
|
||||
# plugin_test(plugin='crazy_functions.命令行助手->命令行助手', main_input='查看当前的docker容器列表')
|
||||
|
||||
plugin_test(plugin='crazy_functions.解析项目源代码->解析一个Python项目', main_input="crazy_functions/test_project/python/dqn")
|
||||
# plugin_test(plugin='crazy_functions.解析项目源代码->解析一个Python项目', main_input="crazy_functions/test_project/python/dqn")
|
||||
|
||||
# plugin_test(plugin='crazy_functions.解析项目源代码->解析一个C项目', main_input="crazy_functions/test_project/cpp/cppipc")
|
||||
|
||||
@ -19,7 +19,7 @@ if __name__ == "__main__":
|
||||
|
||||
# plugin_test(plugin='crazy_functions.批量Markdown翻译->Markdown中译英', main_input="README.md")
|
||||
|
||||
# plugin_test(plugin='crazy_functions.批量翻译PDF文档_多线程->批量翻译PDF文档', main_input="crazy_functions/test_project/pdf_and_word")
|
||||
plugin_test(plugin='crazy_functions.批量翻译PDF文档_多线程->批量翻译PDF文档', main_input='crazy_functions/test_project/pdf_and_word/aaai.pdf')
|
||||
|
||||
# plugin_test(plugin='crazy_functions.谷歌检索小助手->谷歌检索小助手', main_input="https://scholar.google.com/scholar?hl=en&as_sdt=0%2C5&q=auto+reinforcement+learning&btnG=")
|
||||
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user