cont_medium_4_256
This model is a fine-tuned version of sberbank-ai/rugpt3medium_based_on_gpt2 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 3.7602
- Accuracy: 0.3436
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: 1.5e-05
- train_batch_size: 6
- eval_batch_size: 12
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 4.0
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 3.4091 | 1.0 | 160 | 3.7238 | 0.3472 |
| 3.2338 | 2.0 | 320 | 3.7393 | 0.3446 |
| 3.1684 | 3.0 | 480 | 3.7518 | 0.3444 |
| 3.0989 | 4.0 | 640 | 3.7602 | 0.3436 |
Framework versions
- Transformers 4.23.1
- Pytorch 1.8.1+cu111
- Datasets 2.5.2
- Tokenizers 0.13.1