20split_dataset_version4

This model is a fine-tuned version of distilbert-base-uncased on the None dataset. It achieves the following results on the evaluation set:

• Loss: 3.1060

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: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 16

Training results

Training Loss	Epoch	Step	Validation Loss
3.4213	1.0	313	3.2276
3.2295	2.0	626	3.2054
3.0262	3.0	939	3.1718
2.9608	4.0	1252	3.1649
2.8715	5.0	1565	3.1499
2.7614	6.0	1878	3.1471
2.6688	7.0	2191	3.1280
2.6582	8.0	2504	3.1348
2.5296	9.0	2817	3.1149
2.5086	10.0	3130	3.1200
2.4567	11.0	3443	3.1020
2.4052	12.0	3756	3.1257
2.3513	13.0	4069	3.1222
2.3561	14.0	4382	3.1102
2.3574	15.0	4695	3.1027
2.3192	16.0	5008	3.1060

Framework versions

• Transformers 4.21.0
• Pytorch 1.12.0+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1