模型:

philschmid/distilroberta-base-ner-conll2003

英文

distilroberta-base-ner-conll2003

该模型是在conll2003数据集上对 distilroberta-base 进行微调的版本。

评估F1-Score: 95.29(CoNLL-03)测试F1-Score: 90.74(CoNLL-03)

评估F1-Score: 95.29(CoNLL++ / CoNLL-03修正)测试F1-Score: 92.23(CoNLL++ / CoNLL-03修正)

模型使用

from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline

tokenizer = AutoTokenizer.from_pretrained("philschmid/distilroberta-base-ner-conll2003")
model = AutoModelForTokenClassification.from_pretrained("philschmid/distilroberta-base-ner-conll2003")

nlp = pipeline("ner", model=model, tokenizer=tokenizer, grouped_entities=True)
example = "My name is Philipp and live in Germany"

nlp(example)

训练过程

训练超参数

在训练期间使用了以下超参数:

  • learning_rate: 4.9902376275441704e-05
  • train_batch_size: 32
  • eval_batch_size: 16
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 6.0
  • mixed_precision_training: Native AMP

训练结果

CoNNL2003

在评估集上取得以下结果:

  • Loss: 0.0583
  • Precision: 0.9493
  • Recall: 0.9566
  • F1: 0.9529
  • Accuracy: 0.9883

在测试集上取得以下结果:

  • Loss: 0.2025
  • Precision: 0.8999
  • Recall: 0.915
  • F1: 0.9074
  • Accuracy: 0.9741
CoNNL++ / CoNLL2003修正

在评估集上取得以下结果:

  • Loss: 0.0567
  • Precision: 0.9493
  • Recall: 0.9566
  • F1: 0.9529
  • Accuracy: 0.9883

在测试集上取得以下结果:

  • Loss: 0.1359
  • Precision: 0.92
  • Recall: 0.9245
  • F1: 0.9223
  • Accuracy: 0.9785

框架版本

  • Transformers 4.6.1
  • Pytorch 1.8.1+cu101
  • Datasets 1.6.2
  • Tokenizers 0.10.2