DeBERTa-finetuned-SNLI2

This model is a fine-tuned version of gyeoldere/test_trainer on the snli dataset.

Test_trainer model is a fine-tuned version of microsoft/deberta-base on the snli dataset.

This model achieves the following results on the evaluation set:

• NLI accuracy : 0.86
• MLM accuracy : 0.68

Model description

This model fine-tuned to perform 2 tasks simultaneously; NLI task and MLM task.

Output vector of DeBERTa processed through two different fc layer to predict. I used layer structure introduced in BERT paper, which is implemented on huggingface transformers; DebertaForTokenClassification and DebertaForMaskedLM. [https://huggingface.co/docs/transformers/index]

BinaryCrossEntrophyLoss are used for each class, and two losses are added to obtain final loss final_loss = MLM_loss + NLI_loss

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: 128
• eval_batch_size: 128
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Framework versions

• Transformers 4.26.0
• Pytorch 1.13.1+cu116
• Datasets 2.9.0
• Tokenizers 0.13.2