app.py

import gradio as gr
import os
import shutil
from pypdf import PdfReader
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import fitz

TOKENIZER_REPO = "MediaTek-Research/Breeze-7B-Instruct-v1_0"
tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_REPO,local_files_only=False,use_fast=True)
tran_hints = "请将以下的文字转为繁体:"
start_flag="<s>"
end_flag="</s>"
model = AutoModelForCausalLM.from_pretrained(
    TOKENIZER_REPO,
    device_map="auto",
    local_files_only=False,
    torch_dtype=torch.bfloat16
)

def generate(text):
    chat_data = []
    text = text.strip()
    if text:
       chat_data.append({"role": "user", "content": text})
    achat=tokenizer.apply_chat_template(chat_data,return_tensors="pt")
    #achat=tokenizer.encode(chat_data,return_tensors="pt",max_length=2048)
    outputs = model.generate(achat,
                         max_new_tokens=2048,
                         top_p=0.01,
                         top_k=85,
                         repetition_penalty=1.1,
                         temperature=0)

    return tokenizer.decode(outputs[0])

def tran_txt(input_txt):
  data_txt=tran_hints+"\n"+input_txt.strip()
  tran_result=generate(data_txt)
  print("tran_result="+tran_result)
#  tran_result=tran_result.strip()
#  index=tran_result.find(start_flag)
#  if index>=0:
#     tran_result=tran_result[len(start_flag):]
#     tran_result=tran_result.strip()
#     c_index=tran_result.find(data_txt)
#     if c_index>=0:
#        tran_result=tran_result[len(data_txt):]
#     e_index=tran_result.find(end_flag)
#     if e_index>=0:
#        tran_result=tran_result[0:e_index]
  return tran_result

def exec_tran(file):
  temp_file=upload_file(file)
  page_texts=read_paragraphs(temp_file)
  temp_result_file=file;
  file_index=temp_result_file.index('.pdf')
  if file_index!=-1:
    temp_result_file=temp_result_file[0:file_index]
    temp_result_file=temp_result_file+"_result.txt"
  else :
    temp_result_file=temp_result_file+"_result.txt"
  tran_file_name=file.name
  with open(temp_result_file,'w') as fw:
    tran_result=tran_txt(tran_hints)
 #   print(tran_result+"\n")
    for page_content in page_texts:
      #lines=page_content.split('\n')
      #for line_content in lines:
          #print("input="+line_content)
          tran_result=tran_txt(page_content)
 #         print("result="+tran_result)
          fw.write(tran_result+"\n")
  return temp_result_file

def upload_file(file):
  UPLOAD_FOLDER="./data"
  if not os.path.exists(UPLOAD_FOLDER):
    os.mkdir(UPLOAD_FOLDER)
  return shutil.copy(file,UPLOAD_FOLDER)

def read_paragraphs(pdf_path):
    document = fitz.open(pdf_path)
    paragraphs = []

    for page in document:
        text = page.get_text("paragraphs")
        para_list = text.split('。')
        paragraphs.extend([para for para in para_list if para.strip()])
    document.close()
    return paragraphs

def load_pdf_pages(filename):
  page_texts=[]
  reader = PdfReader(filename)
  for page in reader.pages:
     page_texts.append(page.extract_text())
  return page_texts

def exec_translate(file):
  upload_file(file)
  page_texts=load_pdf_pages(file.name)

with gr.Blocks() as app:
   file_output=gr.File()
   upload_button=gr.UploadButton("上传pdf文件",file_types=["pdf"],file_count="single")
   upload_button.upload(exec_tran,upload_button,file_output)

app.launch()