license | tags | model-index | ||||
---|---|---|---|---|---|---|
apache-2.0 |
|
|
This model is a fine-tuned version of prikarsartam/Chatalet on an unknown dataset. It achieves the following results on the evaluation set:
- Train Loss: 2.6154
- Validation Loss: 2.4298
- Train Rouge1: 29.4609
- Train Rouge2: 8.3437
- Train Rougel: 23.0867
- Train Rougelsum: 23.0929
- Train Gen Len: 18.8153
- Epoch: 0
A Seq2Seq model based on Keras model structure with the purpose of extreme-summarisation of any given text of arbitrary inputs; the further plan is to integrate 'multilabel' text classification and 'allure-filter' to enhance performability
Trained on Custom dataset from BBC News Data
It has been trained with 1 epoch with train_loss of 2.6% and will be improved with larger datasets and greated epochs
The following hyperparameters were used during training:
- optimizer: {'name': 'AdamWeightDecay', 'learning_rate': 2e-06, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-07, 'amsgrad': False, 'weight_decay_rate': 0.01}
- training_precision: float32
Train Loss | Validation Loss | Train Rouge1 | Train Rouge2 | Train Rougel | Train Rougelsum | Train Gen Len | Epoch |
---|---|---|---|---|---|---|---|
2.6154 | 2.4298 | 29.4609 | 8.3437 | 23.0867 | 23.0929 | 18.8153 | 0 |
- Transformers 4.21.3
- TensorFlow 2.8.2
- Datasets 2.4.0
- Tokenizers 0.12.1