BERT_full-6_tokenized

This model is a fine-tuned version of armheb/DNA_bert_6 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0362

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: 2e-05
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.0775	1.0	284	0.0411
0.0428	2.0	568	0.0393
0.0395	3.0	852	0.0396
0.0395	4.0	1136	0.0374
0.0379	5.0	1420	0.0379
0.037	6.0	1704	0.0399
0.0368	7.0	1988	0.0382
0.0378	8.0	2272	0.0378
0.0365	9.0	2556	0.0362
0.0374	10.0	2840	0.0359
0.0372	11.0	3124	0.0373
0.0358	12.0	3408	0.0378
0.0361	13.0	3692	0.0385
0.0364	14.0	3976	0.0383
0.035	15.0	4260	0.0376
0.035	16.0	4544	0.0376
0.036	17.0	4828	0.0388
0.0365	18.0	5112	0.0372
0.0355	19.0	5396	0.0363
0.0349	20.0	5680	0.0378
0.0345	21.0	5964	0.0377
0.0349	22.0	6248	0.0372
0.035	23.0	6532	0.0374
0.0351	24.0	6816	0.0379
0.0351	25.0	7100	0.0374

Framework versions

• Transformers 4.23.1
• Pytorch 1.12.1+cu113
• Datasets 2.6.1
• Tokenizers 0.13.1