CNEC_1_1_Supertypes_slavicbert

This model is a fine-tuned version of DeepPavlov/bert-base-bg-cs-pl-ru-cased on the cnec dataset. It achieves the following results on the evaluation set:

• Loss: 0.2993
• Precision: 0.8427
• Recall: 0.8811
• F1: 0.8615
• Accuracy: 0.9511

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 25

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.4662	1.7	500	0.2442	0.7608	0.8311	0.7944	0.9353
0.2083	3.4	1000	0.2039	0.8150	0.8744	0.8437	0.9467
0.1504	5.1	1500	0.1902	0.8234	0.8740	0.8480	0.9517
0.11	6.8	2000	0.2027	0.8328	0.8762	0.8539	0.9519
0.0883	8.5	2500	0.2176	0.8361	0.8820	0.8584	0.9509
0.0708	10.2	3000	0.2297	0.8405	0.8828	0.8611	0.9510
0.0615	11.9	3500	0.2429	0.8361	0.8793	0.8571	0.9519
0.0471	13.61	4000	0.2546	0.8340	0.8775	0.8552	0.9504
0.0428	15.31	4500	0.2718	0.8440	0.8775	0.8604	0.9495
0.0358	17.01	5000	0.2730	0.8401	0.8758	0.8576	0.9502
0.0325	18.71	5500	0.2793	0.8421	0.8815	0.8613	0.9501
0.0277	20.41	6000	0.2984	0.8446	0.8842	0.8639	0.9504
0.0245	22.11	6500	0.2987	0.8454	0.8802	0.8625	0.9507
0.0224	23.81	7000	0.2993	0.8427	0.8811	0.8615	0.9511

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0