BertjeWDialDataALLQonly05

This model is a fine-tuned version of GroNLP/bert-base-dutch-cased on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.3921

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 12.0

Training results

Training Loss	Epoch	Step	Validation Loss
2.9349	1.0	871	2.9642
2.9261	2.0	1742	2.9243
2.8409	3.0	2613	2.8895
2.7308	4.0	3484	2.8394
2.6042	5.0	4355	2.7703
2.4671	6.0	5226	2.7522
2.3481	7.0	6097	2.6339
2.2493	8.0	6968	2.6224
2.1233	9.0	7839	2.5637
2.0194	10.0	8710	2.4896
1.9178	11.0	9581	2.4689
1.8588	12.0	10452	2.4663

Framework versions

• Transformers 4.13.0.dev0
• Pytorch 1.10.0
• Datasets 1.16.1
• Tokenizers 0.10.3