LiLT-SER-FR

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: 2.1399
• Precision: 0.6810
• Recall: 0.7606
• F1: 0.7186
• Accuracy: 0.7894

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
0.084	9.8	500	1.0721	0.6674	0.6029	0.6335	0.7616
0.0351	19.61	1000	1.5259	0.6832	0.7148	0.6986	0.7849
0.0017	29.41	1500	1.5810	0.6707	0.7207	0.6948	0.7753
0.0126	39.22	2000	1.8551	0.7075	0.7021	0.7048	0.7817
0.0006	49.02	2500	1.8431	0.6288	0.7539	0.6857	0.7837
0.0284	58.82	3000	1.8819	0.6434	0.7362	0.6867	0.7759
0.0001	68.63	3500	1.7968	0.7088	0.7005	0.7046	0.7824
0.0001	78.43	4000	1.8480	0.6860	0.7324	0.7084	0.7875
0.0	88.24	4500	1.9706	0.6781	0.7257	0.7011	0.7820
0.0004	98.04	5000	2.0033	0.6974	0.7202	0.7086	0.7841
0.0001	107.84	5500	2.0152	0.6790	0.7333	0.7051	0.7831
0.0	117.65	6000	1.9981	0.6968	0.7350	0.7154	0.7763
0.0	127.45	6500	2.0012	0.6844	0.7425	0.7123	0.7799
0.0001	137.25	7000	2.0329	0.6578	0.7383	0.6957	0.7808
0.0	147.06	7500	2.1239	0.6882	0.7362	0.7114	0.7846
0.0	156.86	8000	2.0940	0.6886	0.7387	0.7128	0.7878
0.0	166.67	8500	2.1262	0.6954	0.7463	0.7200	0.7863
0.0	176.47	9000	2.1399	0.6810	0.7606	0.7186	0.7894
0.0	186.27	9500	2.1534	0.6844	0.7535	0.7173	0.7887
0.0	196.08	10000	2.1526	0.6848	0.7514	0.7165	0.7892

Framework versions

• Transformers 4.38.2
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.1