opd_bert_spam_ru

This model is a fine-tuned version of ai-forever/ruBert-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1034
• F1: 0.9872
• Accuracy: 0.9872

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: 8
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 70
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	F1	Accuracy
0.1094	0.5102	100	0.2559	0.9587	0.9591
0.1566	1.0204	200	0.0769	0.9847	0.9847
0.0352	1.5306	300	0.0860	0.9821	0.9821
0.0571	2.0408	400	0.0908	0.9846	0.9847
0.0256	2.5510	500	0.1034	0.9872	0.9872
0.0178	3.0612	600	0.1416	0.9770	0.9770
0.0098	3.5714	700	0.1461	0.9821	0.9821
0.0097	4.0816	800	0.1618	0.9770	0.9770
0.0003	4.5918	900	0.1651	0.9795	0.9795

Framework versions

• Transformers 4.40.2
• Pytorch 2.2.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1