Edit model card

kobigbird-bert-base-finetuned-klue-goorm-q-a-task

This model is a fine-tuned version of ToToKr/kobigbird-bert-base-finetuned-klue on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 1.2115

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: 4
  • eval_batch_size: 4
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 20

Training results

Training Loss Epoch Step Validation Loss
1.6159 0.09 500 1.7522
1.554 0.17 1000 1.5953
1.4493 0.26 1500 1.3769
1.4051 0.35 2000 1.3746
1.3251 0.43 2500 1.5049
1.2855 0.52 3000 1.1733
1.2226 0.6 3500 1.1538
1.1907 0.69 4000 1.1470
1.1655 0.78 4500 1.0759
1.1411 0.86 5000 1.0676
1.0752 0.95 5500 0.9894
0.9389 1.04 6000 1.2020
0.8457 1.12 6500 1.1004
0.7977 1.21 7000 1.1397
0.818 1.29 7500 1.2960
0.8142 1.38 8000 1.2115

Framework versions

  • Transformers 4.18.0
  • Pytorch 1.10.0+cu111
  • Datasets 2.1.0
  • Tokenizers 0.12.1
Downloads last month
2
Hosted inference API
Question Answering
Examples
Examples
This model can be loaded on the Inference API on-demand.