metadata

license: apache-2.0
tags:
  - generated_from_trainer
datasets:
  - dair-ai/emotion
metrics:
  - accuracy
  - f1
model-index:
  - name: distilbert-base-uncased-finetuned-emotion
    results:
      - task:
          name: Text Classification
          type: text-classification
        dataset:
          name: emotion
          type: emotion
          config: split
          split: validation
          args: split
        metrics:
          - name: Accuracy
            type: accuracy
            value: 0.9375
          - name: F1
            type: f1
            value: 0.937890467332837
language:
  - en

distilbert-base-uncased-finetuned-emotion

This model is a fine-tuned version of distilbert-base-uncased on the emotion dataset. It achieves the following results on the evaluation set:

Loss: 0.1448
Accuracy: 0.9375
F1: 0.9379

The notebook used to fine-tune this model may be found HERE.

Model description

DistilBERT is a transformers model, smaller and faster than BERT, which was pretrained on the same corpus in a self-supervised fashion, using the BERT base model as a teacher. This means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts using the BERT base model. More precisely, it was pretrained with three objectives:

Distillation loss: the model was trained to return the same probabilities as the BERT base model.
Masked language modeling (MLM): this is part of the original training loss of the BERT base model. When taking a sentence, the model randomly masks 15% of the words in the input then run the entire masked sentence through the model and has to predict the masked words. This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the sentence.
Cosine embedding loss: the model was also trained to generate hidden states as close as possible as the BERT base model.

This way, the model learns the same inner representation of the English language than its teacher model, while being faster for inference or downstream tasks.

Intended uses & limitations

Emotion is a dataset of English Twitter messages with six basic emotions: anger, fear, joy, love, sadness, and surprise. This dataset was developed for the paper entitled "CARER: Contextualized Affect Representations for Emotion Recognition" (Saravia et al.) through noisy labels, annotated via distant supervision as in the paper"Twitter sentiment classification using distant supervision" (Go et al).

The DistilBERT model was fine-tuned to this dataset, allowing for the classification of sentences into one of the six basic emotions (anger, fear, joy, love, sadness, and surprise).

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-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
0.5337	1.0	250	0.1992	0.927	0.9262
0.1405	2.0	500	0.1448	0.9375	0.9379

Framework versions

Transformers 4.30.2
Pytorch 2.0.0
Datasets 2.1.0
Tokenizers 0.13.3