metadata

license: apache-2.0
tags:
  - generated_from_trainer
datasets:
  - clinc_oos
metrics:
  - accuracy
model-index:
  - name: distilbert-base-uncased-finetuned-clinc
    results:
      - task:
          name: Text Classification
          type: text-classification
        dataset:
          name: clinc_oos
          type: clinc_oos
          args: plus
        metrics:
          - name: Accuracy
            type: accuracy
            value: 0.9180645161290323

distilbert-base-uncased-finetuned-clinc

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

Loss: 0.7719
Accuracy: 0.9181

Model description

This is an initial example of knowledge-distillation where the student loss is all cross-entropy loss $L_{CE}$ of the ground-truth labels and none of the distillation loss $L_{KD}$ .

Intended uses & limitations

More information needed

Training and evaluation data

The training and evaluation data come straight from the train and validation splits in the clinc_oos dataset, respectively; and tokenized using the distilbert-base-uncased tokenization.

Training procedure

Please see page 224 in Chapter 8: Making Transformers Efficient in Production, Natural Language Processing with Transformers, May 2022.

Training hyperparameters

The following hyperparameters were used during training:

alpha: 1.0
temperature: 2.0
learning_rate: 2e-05
train_batch_size: 48
eval_batch_size: 48
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	Accuracy
No log	1.0	318	3.2882	0.7426
3.7861	2.0	636	1.8744	0.8381
3.7861	3.0	954	1.1567	0.8958
1.6922	4.0	1272	0.8569	0.9132
0.9055	5.0	1590	0.7719	0.9181

Framework versions

Transformers 4.16.2
Pytorch 2.1.2+cu121
Datasets 1.16.1
Tokenizers 0.15.1