hmBERT-CoNLL-cp1

This model is a fine-tuned version of dbmdz/bert-base-historic-multilingual-cased on the conll2003 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0710
• Precision: 0.8690
• Recall: 0.8888
• F1: 0.8788
• Accuracy: 0.9810

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: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	0.06	25	0.4115	0.3593	0.3708	0.3649	0.9002
No log	0.11	50	0.2263	0.6360	0.6898	0.6618	0.9456
No log	0.17	75	0.1660	0.7250	0.7582	0.7412	0.9564
No log	0.23	100	0.1520	0.7432	0.7775	0.7600	0.9597
No log	0.28	125	0.1343	0.7683	0.8103	0.7888	0.9645
No log	0.34	150	0.1252	0.7973	0.8230	0.8099	0.9691
No log	0.4	175	0.1021	0.8118	0.8398	0.8255	0.9724
No log	0.46	200	0.1056	0.8153	0.8411	0.8280	0.9727
No log	0.51	225	0.0872	0.8331	0.8612	0.8469	0.9755
No log	0.57	250	0.1055	0.8226	0.8418	0.8321	0.9725
No log	0.63	275	0.0921	0.8605	0.8640	0.8623	0.9767
No log	0.68	300	0.0824	0.8600	0.8787	0.8692	0.9788
No log	0.74	325	0.0834	0.8530	0.8771	0.8649	0.9787
No log	0.8	350	0.0758	0.8646	0.8876	0.8759	0.9800
No log	0.85	375	0.0727	0.8705	0.8866	0.8784	0.9810
No log	0.91	400	0.0734	0.8717	0.8899	0.8807	0.9811
No log	0.97	425	0.0713	0.8683	0.8889	0.8785	0.9810

Framework versions

• Transformers 4.20.1
• Pytorch 1.12.0
• Datasets 2.4.0
• Tokenizers 0.12.1