This is only a small scale fine-tuning of the standfordnlp/imbd datasets. Only 1000 rows of the unsupervised dataset is used for training. The exercise is carried on Google Colab - T4 gpu.

Training and evaluation data

1000 rows from the standfordnlp/imbd datasets.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

optimizer: {'name': 'AdamWeightDecay', 'learning_rate': {'module': 'transformers.optimization_tf', 'class_name': 'WarmUp', 'config': {'initial_learning_rate': 2e-05, 'decay_schedule_fn': {'module': 'keras.optimizers.schedules', 'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 2e-05, 'decay_steps': -969, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}, 'registered_name': None}, 'warmup_steps': 1000, 'power': 1.0, 'name': None}, 'registered_name': 'WarmUp'}, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-08, 'amsgrad': False, 'weight_decay_rate': 0.01}
training_precision: mixed_float16

Training results

Train Loss	Validation Loss	Epoch
3.2484	3.2338	0
3.0821	2.8758	1
2.9373	2.9930	2

Framework versions

Transformers 4.40.2
TensorFlow 2.15.0
Datasets 2.19.1
Tokenizers 0.19.1