twitter-data-bert-base-multilingual-uncased-hindi-only-memes
This model is a fine-tuned version of bert-base-multilingual-uncased on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.5614
- Accuracy: 0.9031
- Precision: 0.9064
- Recall: 0.9057
- F1: 0.9060
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: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 6
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Precision | Recall | F1 |
|---|---|---|---|---|---|---|---|
| 0.5408 | 1.0 | 511 | 0.4798 | 0.7974 | 0.8447 | 0.8049 | 0.7940 |
| 0.3117 | 2.0 | 1022 | 0.3576 | 0.8844 | 0.8875 | 0.8882 | 0.8869 |
| 0.2019 | 3.0 | 1533 | 0.3401 | 0.9020 | 0.9076 | 0.9047 | 0.9052 |
| 0.1364 | 4.0 | 2044 | 0.4519 | 0.8888 | 0.8936 | 0.8921 | 0.8923 |
| 0.0767 | 5.0 | 2555 | 0.5251 | 0.8987 | 0.9024 | 0.9016 | 0.9019 |
| 0.0433 | 6.0 | 3066 | 0.5614 | 0.9031 | 0.9064 | 0.9057 | 0.9060 |
Framework versions
- Transformers 4.24.0
- Pytorch 1.12.1+cu113
- Datasets 2.6.1
- Tokenizers 0.13.1