san_nli_2409_1325
This model is a fine-tuned version of svalabs/gbert-large-zeroshot-nli on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.3856
- F1: 0.9219
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: 5e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 4
Training results
Training Loss | Epoch | Step | Validation Loss | F1 |
---|---|---|---|---|
No log | 0.93 | 10 | 0.2410 | 0.9219 |
No log | 1.93 | 20 | 0.5240 | 0.9149 |
No log | 2.93 | 30 | 0.4756 | 0.9219 |
No log | 3.93 | 40 | 0.3856 | 0.9219 |
Framework versions
- Transformers 4.22.2
- Pytorch 1.12.1+cu113
- Datasets 2.5.1
- Tokenizers 0.12.1
- Downloads last month
- 8