bertbek-ner-uznews

This model is a fine-tuned version of elmurod1202/bertbek-news-big-cased on UzNER dataset. It achieves the following results on the evaluation set:

• Loss: 0.1255
• Date: {'precision': 0.8585858585858586, 'recall': 0.9205776173285198, 'f1': 0.8885017421602788, 'number': 277}
• Location: {'precision': 0.7639344262295082, 'recall': 0.7871621621621622, 'f1': 0.7753743760399334, 'number': 296}
• Misc: {'precision': 0.5661375661375662, 'recall': 0.622093023255814, 'f1': 0.5927977839335179, 'number': 172}
• Org: {'precision': 0.5685279187817259, 'recall': 0.5743589743589743, 'f1': 0.5714285714285714, 'number': 195}
• Person: {'precision': 0.8644501278772379, 'recall': 0.9415041782729805, 'f1': 0.9013333333333333, 'number': 359}
• Time: {'precision': 0.14285714285714285, 'recall': 0.1, 'f1': 0.11764705882352941, 'number': 10}
• Overall Precision: 0.7547
• Overall Recall: 0.7991
• Overall F1: 0.7763
• Overall Accuracy: 0.9616

Model description

BERTbek model fine-tuned for the NER task

Training and evaluation data

UzNER dataset

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Date	Location	Misc	Org	Person	Time	Overall Precision	Overall Recall	Overall F1	Overall Accuracy
No log	1.0	159	0.2034	{'precision': 0.7120743034055728, 'recall': 0.8303249097472925, 'f1': 0.7666666666666667, 'number': 277}	{'precision': 0.7021943573667712, 'recall': 0.7567567567567568, 'f1': 0.7284552845528456, 'number': 296}	{'precision': 0.43902439024390244, 'recall': 0.20930232558139536, 'f1': 0.28346456692913385, 'number': 172}	{'precision': 0.4134078212290503, 'recall': 0.37948717948717947, 'f1': 0.3957219251336898, 'number': 195}	{'precision': 0.6337349397590362, 'recall': 0.7325905292479109, 'f1': 0.6795865633074936, 'number': 359}	{'precision': 0.0, 'recall': 0.0, 'f1': 0.0, 'number': 10}	0.6256	0.6318	0.6287	0.9397
No log	2.0	318	0.1454	{'precision': 0.839041095890411, 'recall': 0.8844765342960289, 'f1': 0.8611599297012302, 'number': 277}	{'precision': 0.7539936102236422, 'recall': 0.7972972972972973, 'f1': 0.7750410509031198, 'number': 296}	{'precision': 0.5465116279069767, 'recall': 0.5465116279069767, 'f1': 0.5465116279069767, 'number': 172}	{'precision': 0.5721649484536082, 'recall': 0.5692307692307692, 'f1': 0.570694087403599, 'number': 195}	{'precision': 0.7985257985257985, 'recall': 0.9052924791086351, 'f1': 0.8485639686684072, 'number': 359}	{'precision': 0.0, 'recall': 0.0, 'f1': 0.0, 'number': 10}	0.7305	0.7723	0.7508	0.9586
No log	3.0	477	0.1333	{'precision': 0.8246753246753247, 'recall': 0.9169675090252708, 'f1': 0.8683760683760684, 'number': 277}	{'precision': 0.7005988023952096, 'recall': 0.7905405405405406, 'f1': 0.7428571428571429, 'number': 296}	{'precision': 0.5204081632653061, 'recall': 0.5930232558139535, 'f1': 0.5543478260869567, 'number': 172}	{'precision': 0.5784313725490197, 'recall': 0.6051282051282051, 'f1': 0.5914786967418547, 'number': 195}	{'precision': 0.850253807106599, 'recall': 0.9331476323119777, 'f1': 0.8897742363877823, 'number': 359}	{'precision': 0.14285714285714285, 'recall': 0.1, 'f1': 0.11764705882352941, 'number': 10}	0.7235	0.7976	0.7587	0.9583
0.2001	4.0	636	0.1255	{'precision': 0.8585858585858586, 'recall': 0.9205776173285198, 'f1': 0.8885017421602788, 'number': 277}	{'precision': 0.7639344262295082, 'recall': 0.7871621621621622, 'f1': 0.7753743760399334, 'number': 296}	{'precision': 0.5661375661375662, 'recall': 0.622093023255814, 'f1': 0.5927977839335179, 'number': 172}	{'precision': 0.5685279187817259, 'recall': 0.5743589743589743, 'f1': 0.5714285714285714, 'number': 195}	{'precision': 0.8644501278772379, 'recall': 0.9415041782729805, 'f1': 0.9013333333333333, 'number': 359}	{'precision': 0.14285714285714285, 'recall': 0.1, 'f1': 0.11764705882352941, 'number': 10}	0.7547	0.7991	0.7763	0.9616

Framework versions

• Transformers 4.50.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1