LiLT-SER-PT

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: 2.1403
• Precision: 0.6998
• Recall: 0.7551
• F1: 0.7264
• Accuracy: 0.7710

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	Accuracy	F1	Validation Loss	Precision	Recall
0.0838	8.47	500	0.7697	0.6542	1.0006	0.6081	0.7078
0.0366	16.95	1000	0.7606	0.6795	1.4063	0.6533	0.7078
0.0173	25.42	1500	0.7848	0.7047	1.4681	0.6752	0.7369
0.0036	33.9	2000	0.7706	0.7003	1.6267	0.6577	0.7487
0.0023	42.37	2500	1.6728	0.6839	0.7172	0.7002	0.7698
0.0001	50.85	3000	1.6210	0.6742	0.7493	0.7098	0.7941
0.0001	59.32	3500	1.6883	0.6962	0.7505	0.7223	0.7929
0.0007	67.8	4000	1.8709	0.6730	0.7590	0.7134	0.7811
0.0003	76.27	4500	1.9387	0.6884	0.7151	0.7015	0.7690
0.0034	84.75	5000	1.8042	0.6927	0.7554	0.7227	0.7787
0.0	93.22	5500	2.0395	0.6954	0.7596	0.7261	0.7527
0.0003	101.69	6000	1.9295	0.6861	0.7511	0.7172	0.7790
0.0001	110.17	6500	1.9690	0.6813	0.7611	0.7190	0.7694
0.0	118.64	7000	1.9217	0.6974	0.7520	0.7237	0.7754
0.0001	127.12	7500	2.0703	0.6885	0.7536	0.7196	0.7694
0.0002	135.59	8000	2.0438	0.6915	0.7635	0.7258	0.7770
0.0	144.07	8500	2.0429	0.6980	0.7599	0.7276	0.7782
0.0	152.54	9000	2.1403	0.6998	0.7551	0.7264	0.7710
0.0	161.02	9500	2.1786	0.6986	0.7578	0.7270	0.7726
0.0	169.49	10000	2.1782	0.6965	0.7560	0.7250	0.7721

Framework versions

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