<!-- 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. -->
distilbert-base-uncased-finetuned-emotion
This model is a fine-tuned version of distilbert-base-uncased on SetFit/emotion. It achieves the following results on the evaluation set:
- Loss: 0.2276
- Accuracy: 0.921
- F1: 0.9209
Model description
This model follows chapter 2 of https://github.com/nlp-with-transformers/notebooks. A few things that were changed from the original notebook:
- the emotion dataset has moved to SetFit/emotion https://github.com/nlp-with-transformers/notebooks/issues/77
- the new dataset doesn't have ClassLabel feature so needed to change int2str method https://github.com/nlp-with-transformers/notebooks/issues/77
- made the label names on inference API human-readable with https://discuss.huggingface.co/t/change-label-names-on-inference-api/3063/3
- function to inspect dataset for existence of certain strings
Intended uses & limitations
Training and evaluation data
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: linear
- num_epochs: 2
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 |
|---|---|---|---|---|---|
| 0.8732 | 1.0 | 250 | 0.3279 | 0.9055 | 0.9037 |
| 0.259 | 2.0 | 500 | 0.2276 | 0.921 | 0.9209 |
Framework versions
- Transformers 4.13.0
- Pytorch 1.13.0+cu116
- Datasets 1.16.1
- Tokenizers 0.10.3
