LiLT-SER-JA

This model is a fine-tuned version of nielsr/lilt-xlm-roberta-base on the xfun dataset. It achieves the following results on the evaluation set:

• Loss: 2.3482
• Precision: 0.7244
• Recall: 0.8755
• F1: 0.7928
• Accuracy: 0.7835

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: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.0726	10.2	500	1.0347	0.6824	0.8359	0.7514	0.7829
0.0015	20.41	1000	1.6415	0.6828	0.8808	0.7692	0.7700
0.0062	30.61	1500	1.7000	0.7063	0.8427	0.7685	0.7828
0.0145	40.82	2000	1.9098	0.6979	0.8885	0.7817	0.7729
0.0014	51.02	2500	1.6868	0.7117	0.8509	0.7751	0.7859
0.0009	61.22	3000	1.8930	0.7087	0.8441	0.7705	0.7782
0.0001	71.43	3500	2.0325	0.7217	0.8736	0.7904	0.7845
0.0006	81.63	4000	1.8854	0.7032	0.8769	0.7805	0.7904
0.0001	91.84	4500	2.2205	0.6977	0.8721	0.7752	0.7577
0.0002	102.04	5000	2.1731	0.7090	0.8702	0.7814	0.7786
0.0	112.24	5500	2.3198	0.7150	0.8707	0.7852	0.7681
0.0003	122.45	6000	1.9680	0.7188	0.8649	0.7851	0.7896
0.0	132.65	6500	2.2202	0.7316	0.8523	0.7873	0.7815
0.0	142.86	7000	2.2800	0.7013	0.8818	0.7813	0.7727
0.0	153.06	7500	2.2149	0.7202	0.8784	0.7915	0.7790
0.0	163.27	8000	2.2384	0.7264	0.8663	0.7902	0.7834
0.0001	173.47	8500	2.2177	0.7269	0.8682	0.7913	0.7842
0.0	183.67	9000	2.2768	0.7333	0.8731	0.7971	0.7872
0.0	193.88	9500	2.2996	0.7344	0.8716	0.7972	0.7878
0.0	204.08	10000	2.3482	0.7244	0.8755	0.7928	0.7835

Framework versions

• Transformers 4.39.1
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.1