Roberta55's picture
update model card README.md
c3d249f
---
license: mit
tags:
- generated_from_trainer
datasets:
- glue
metrics:
- matthews_correlation
model-index:
- name: deberta-base-mnli-finetuned-cola
results:
- task:
name: Text Classification
type: text-classification
dataset:
name: glue
type: glue
args: cola
metrics:
- name: Matthews Correlation
type: matthews_correlation
value: 0.6281691768918801
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# deberta-base-mnli-finetuned-cola
This model is a fine-tuned version of [microsoft/deberta-base-mnli](https://huggingface.co/microsoft/deberta-base-mnli) on the glue dataset.
It achieves the following results on the evaluation set:
- Loss: 0.8205
- Matthews Correlation: 0.6282
## 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: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5
### Training results
| Training Loss | Epoch | Step | Validation Loss | Matthews Correlation |
|:-------------:|:-----:|:----:|:---------------:|:--------------------:|
| 0.4713 | 1.0 | 535 | 0.5110 | 0.5797 |
| 0.2678 | 2.0 | 1070 | 0.6648 | 0.5154 |
| 0.1811 | 3.0 | 1605 | 0.6681 | 0.6121 |
| 0.113 | 4.0 | 2140 | 0.8205 | 0.6282 |
| 0.0831 | 5.0 | 2675 | 1.0413 | 0.6057 |
### Framework versions
- Transformers 4.11.3
- Pytorch 1.9.0+cu111
- Datasets 1.14.0
- Tokenizers 0.10.3