---
metrics:
- accuracy
- f1
- precision
- recall
model-index:
- name: twitter-roberta-base-sentiment-mlm-class
---
<!-- 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. -->
# twitter-roberta-base-sentiment-mlm-class
This model was trained from scratch on an unkown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.3783
- Accuracy: 0.9592
- F1: 0.9553
- Precision: 0.9496
- Recall: 0.9618
## 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: 64
- eval_batch_size: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine_with_restarts
- num_epochs: 20
- mixed_precision_training: Native AMP
- label_smoothing_factor: 0.1
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 | Precision | Recall |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:------:|:---------:|:------:|
| 0.4411 | 1.0 | 1055 | 0.3862 | 0.9528 | 0.9477 | 0.9426 | 0.9536 |
| 0.3612 | 2.0 | 2110 | 0.3730 | 0.9592 | 0.9548 | 0.9546 | 0.9550 |
| 0.3365 | 3.0 | 3165 | 0.3783 | 0.9592 | 0.9553 | 0.9496 | 0.9618 |
### Framework versions
- Transformers 4.6.1
- Pytorch 1.7.0
- Datasets 1.11.0
- Tokenizers 0.10.3