Hierarchical BERT
Collection
Set of BERT models with Hierarchical attention pre-trained on conversational data to process multiple utterances at once • 8 items • Updated
hier-bert-i3-mlm2
This model is a fine-tuned version of HierBert on an English version of OpenSubtitles dataset. It achieves the following results on the evaluation set:
- Loss: 2.2929
- Accuracy: 0.5612
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.001
- train_batch_size: 16
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 4
- total_train_batch_size: 256
- total_eval_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.5
- training_steps: 50000
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 2.9488 | 1.55 | 25000 | 2.7667 | 0.4935 |
| 2.4233 | 3.1 | 50000 | 2.2922 | 0.5612 |
Framework versions
- Transformers 4.31.0
- Pytorch 2.0.1+cu117
- Datasets 2.14.4
- Tokenizers 0.13.3
- Downloads last month
- 17