paddlenlp taskflow

很抱歉，作为语言模型，我无法提供窗口程序编写方面的支持。不过我可以给你提供一些基于PaddleNLP的taskflow实现的代码来提取PDF文档信息，供你参考。 ```python import paddle import paddle.nn as nn from paddle.static import InputSpec from paddlenlp.datasets import load_dataset from paddlenlp.data import Stack, Tuple, Pad from paddlenlp.transformers import BertModel, BertForTokenClassification from paddlenlp.metrics import ChunkEvaluator from paddlenlp.utils.log import logger from paddle.io import DataLoader from typing import List, Tuple import os import fitz import re import tkinter as tk from tkinter import filedialog paddle.set_device('cpu') class PDFReader: def __init__(self): self.model = None self.label_map = { 0: 'O', 1: 'B-NAME', 2: 'I-NAME', 3: 'B-ADDRESS', 4: 'I-ADDRESS', 5: 'B-PHONE', 6: 'I-PHONE', 7: 'B-EMAIL', 8: 'I-EMAIL', 9: 'B-ORGANIZATION', 10: 'I-ORGANIZATION' } self.rev_label_map = {v: k for k, v in self.label_map.items()} def load_model(self, model_path: str): self.model = BertForTokenClassification.from_pretrained( model_path, num_classes=len(self.label_map), dropout=None) def extract_info(self, pdf_path: str) -> List[Tuple[str, str]]: assert self.model is not None, "Please load model first" doc = fitz.open(pdf_path) info_list = [] for page in doc: text = page.get_text() text = re.sub('\s+', ' ', text) tokens = [] for char in text: tokens.append(char) labels = self.predict(tokens) entities = self.extract_entities(tokens, labels) info_list.extend(entities) return info_list def predict(self, tokens: List[str]) -> List[int]: self.model.eval() encoded_inputs = self.model.tokenizer( tokens, return_length=True, is_split_into_words=True) input_ids = paddle.to_tensor(encoded_inputs['input_ids']) token_type_ids = paddle.to_tensor(encoded_inputs['token_type_ids']) seq_len = paddle.to_tensor(encoded_inputs['seq_len']) logits = self.model(input_ids, token_type_ids, seq_len) preds = paddle.argmax(logits, axis=-1) return preds.numpy().tolist()[0] def extract_entities(self, tokens: List[str], labels: List[int]) -> List[Tuple[str, str]]: entities = [] entity_type = None entity_value = "" for i, label in enumerate(labels): if label == self.rev_label_map['O']: if entity_type is not None and entity_value != "": entities.append((entity_type, entity_value)) entity_type = None entity_value = "" else: if entity_type is None: entity_type = self.label_map[label].split('-')[1] entity_value += tokens[i] else: if self.label_map[label].split('-')[1] == entity_type: entity_value += tokens[i] else: entities.append((entity_type, entity_value)) entity_type = self.label_map[label].split('-')[1] entity_value = tokens[i] if entity_type is not None and entity_value != "": entities.append((entity_type, entity_value)) return entities class PDFReaderWindow: def __init__(self): self.reader = PDFReader() self.reader.load_model("path/to/model") self.window = tk.Tk() self.window.title("PDF Reader") self.window.geometry("400x400") self.pdf_path_label = tk.Label(self.window, text="PDF Path:") self.pdf_path_label.pack() self.pdf_path_entry = tk.Entry(self.window, width=50) self.pdf_path_entry.pack() self.choose_pdf_button = tk.Button( self.window, text="Choose PDF", command=self.choose_pdf) self.choose_pdf_button.pack() self.extract_info_button = tk.Button( self.window, text="Extract Info", command=self.extract_info) self.extract_info_button.pack() self.info_text = tk.Text(self.window, height=20) self.info_text.pack() def choose_pdf(self): pdf_path = filedialog.askopenfilename( title="Choose PDF", filetypes=(("PDF Files", "*.pdf"),)) self.pdf_path_entry.delete(0, tk.END) self.pdf_path_entry.insert(0, pdf_path) def extract_info(self): pdf_path = self.pdf_path_entry.get() info_list = self.reader.extract_info(pdf_path) self.info_text.delete('1.0', tk.END) for info in info_list: self.info_text.insert(tk.END, f"{info[0]}: {info[1]}\n") def run(self): self.window.mainloop() if __name__ == '__main__': window = PDFReaderWindow() window.run() ``` 这个程序实现了一个简单的窗口应用程序，用户可以选择一个PDF文件，并点击按钮提取其中的信息。程序使用了PaddleNLP中的一个预训练模型来做实体识别。

这个错误通常是由于PaddleNLP和Taskflow之间的循环依赖导致的。可以尝试升级PaddleNLP和Taskflow的版本，或者使用以下命令解决问题： ``` pip uninstall paddlenlp taskflow pip install paddlenlp -U pip install taskflow -U ``` 这将卸载现有版本并重新安装最新版本。如果问题仍然存在，可以尝试卸载PaddlePaddle和PaddleNLP，然后重新安装它们。