roberta-base-ner-demo-turshilt2

This model is a fine-tuned version of bayartsogt/mongolian-roberta-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1242
• Precision: 0.9296
• Recall: 0.9365
• F1: 0.9330
• Accuracy: 0.9802

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: 3e-05
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.6386	0.9958	119	0.1340	0.7472	0.8012	0.7732	0.9536
0.1096	2.0	239	0.0939	0.8249	0.8791	0.8511	0.9686
0.0647	2.9958	358	0.0893	0.8356	0.8889	0.8614	0.9715
0.0455	4.0	478	0.0963	0.8452	0.8912	0.8676	0.9712
0.0306	4.9958	597	0.0909	0.9234	0.9314	0.9274	0.9795
0.0146	6.0	717	0.1046	0.9235	0.9302	0.9268	0.9789
0.0108	6.9958	836	0.1040	0.9204	0.9311	0.9257	0.9794
0.0079	8.0	956	0.1168	0.9245	0.9309	0.9277	0.9792
0.0063	8.9958	1075	0.1138	0.9258	0.9337	0.9297	0.9800
0.0051	10.0	1195	0.1165	0.9268	0.9330	0.9299	0.9800
0.0047	10.9958	1314	0.1199	0.9261	0.9359	0.9309	0.9803
0.0034	12.0	1434	0.1238	0.9284	0.9358	0.9321	0.9800
0.0027	12.9958	1553	0.1242	0.9267	0.9355	0.9311	0.9800
0.0026	14.0	1673	0.1232	0.9294	0.9365	0.9329	0.9804
0.0023	14.9372	1785	0.1242	0.9296	0.9365	0.9330	0.9802

Framework versions

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