distilbert_classifier_newsgroup
This model is a fine-tuned version of distilbert-base-uncased on the 20 Newsgroups data set that is a collection of approximately 20,000 newsgroup documents, partitioned (nearly) evenly across 20 different newsgroups. It achieves the following results on the evaluation set: loss: 0.5660 - accuracy: 0.8371
Training and evaluation data
- The training set contained 10182 rows with features - text and label.
- The evaluation set contained 7532 rows with features - text and label.
Training procedure
- Setup the model checkpoint as distilbert-base-uncased
- Intialize the starting weights with the model checkpoint and give it the number of labels - i.e., 20 in this case.
- Will be training for 3 epochs
- Using a batch size of 16
Training hyperparameters
The following hyperparameters were used during training:
- optimizer: {'name': 'Adam', 'weight_decay': None, 'clipnorm': None, 'global_clipnorm': None, 'clipvalue': None, 'use_ema': False, 'ema_momentum': 0.99, 'ema_overwrite_frequency': None, 'jit_compile': True, 'is_legacy_optimizer': False, 'learning_rate': {'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 2e-05, 'decay_steps': 1908, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}}, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-08, 'amsgrad': False}
- training_precision: float32
Training results
- Results on the evaluation set: loss: 0.5660 - accuracy: 0.8371
Framework versions
- Transformers 4.28.0
- TensorFlow 2.12.0
- Datasets 2.12.0
- Tokenizers 0.13.3