reformer-finetuned-big_patent-wikipedia-arxiv-16384

This model is a fine-tuned version of robingeibel/reformer-finetuned-big_patent-wikipedia-arxiv-16384 on the wikipedia dataset. It achieves the following results on the evaluation set:

• Loss: 6.5256

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: 2.5e-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
8.0368	1.0	3785	6.7392
6.7992	2.0	7570	6.5576
6.6926	3.0	11355	6.5256

Framework versions

• Transformers 4.20.1
• Pytorch 1.12.0+cu113
• Datasets 2.3.2
• Tokenizers 0.12.1