legal-data-mDeBERTa_V3

This model is a fine-tuned version of MoritzLaurer/mDeBERTa-v3-base-mnli-xnli on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6731
• Accuracy: 0.7634
• Precision: 0.7683
• Recall: 0.7644
• F1: 0.7623
• Ratio: 0.3297

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: 0.005
• train_batch_size: 20
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 40
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.06
• lr_scheduler_warmup_steps: 4
• num_epochs: 15
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1	Ratio
1.4203	0.34	10	1.5822	0.6022	0.6054	0.6046	0.5997	0.3226
1.1177	0.69	20	0.8339	0.7240	0.7270	0.7253	0.7258	0.3262
0.9484	1.03	30	0.7998	0.7168	0.7610	0.7192	0.6951	0.3190
0.9257	1.38	40	0.7183	0.7204	0.7221	0.7220	0.7219	0.3297
0.9529	1.72	50	0.7397	0.6989	0.7022	0.7001	0.6959	0.3297
0.9111	2.07	60	0.6820	0.7204	0.7215	0.7216	0.7188	0.3333
0.9021	2.41	70	0.6832	0.7563	0.7644	0.7570	0.7509	0.3333
0.8849	2.76	80	0.7858	0.7204	0.7365	0.7227	0.7079	0.3297
0.8767	3.1	90	0.8523	0.5520	0.6258	0.5527	0.5677	0.1935
0.9186	3.45	100	0.6877	0.7276	0.7430	0.7283	0.7183	0.3262
0.9127	3.79	110	0.6426	0.7348	0.7398	0.7357	0.7298	0.3333
0.9126	4.14	120	0.7509	0.7348	0.7564	0.7370	0.7215	0.3297
0.8477	4.48	130	0.6818	0.7491	0.7684	0.7497	0.7406	0.3262
0.8747	4.83	140	0.7813	0.6810	0.7704	0.6842	0.6067	0.3262
0.9112	5.17	150	0.7799	0.7204	0.8141	0.7205	0.6686	0.3297
0.8767	5.52	160	0.7959	0.6989	0.8418	0.7021	0.6271	0.3297
0.863	5.86	170	0.7007	0.7240	0.7395	0.7247	0.7139	0.3262
0.9029	6.21	180	0.6524	0.7634	0.7717	0.7642	0.7621	0.3262
0.8427	6.55	190	0.7417	0.7133	0.7374	0.7157	0.6957	0.3262
0.8945	6.9	200	0.7312	0.7527	0.7738	0.7532	0.7437	0.3262
0.8913	7.24	210	0.6410	0.7455	0.7523	0.7473	0.7433	0.3297
0.8848	7.59	220	0.7137	0.7563	0.7585	0.7574	0.7567	0.3297
0.8553	7.93	230	0.6940	0.7599	0.7743	0.7605	0.7530	0.3297
0.8154	8.28	240	0.6460	0.7276	0.7453	0.7298	0.7154	0.3297
0.8842	8.62	250	0.7455	0.7563	0.7694	0.7570	0.7498	0.3297
0.8773	8.97	260	0.7369	0.7348	0.7490	0.7367	0.7291	0.3262
0.8615	9.31	270	0.6577	0.7455	0.7539	0.7464	0.7411	0.3297
0.8664	9.66	280	0.6970	0.7563	0.7631	0.7580	0.7545	0.3297
0.8855	10.0	290	0.7167	0.7204	0.7269	0.7224	0.7169	0.3297
0.8564	10.34	300	0.6808	0.7670	0.7846	0.7676	0.7594	0.3297
0.841	10.69	310	0.6604	0.7455	0.7491	0.7472	0.7455	0.3297
0.8415	11.03	320	0.7150	0.7563	0.7694	0.7570	0.7498	0.3297
0.848	11.38	330	0.6495	0.7670	0.7685	0.7682	0.7680	0.3297
0.8648	11.72	340	0.7094	0.7348	0.7562	0.7369	0.7245	0.3262
0.8465	12.07	350	0.7125	0.7384	0.7758	0.7387	0.7181	0.3262
0.8875	12.41	360	0.6962	0.7563	0.7590	0.7573	0.7564	0.3297
0.8192	12.76	370	0.6496	0.7455	0.7539	0.7464	0.7411	0.3297
0.8089	13.1	380	0.6569	0.7599	0.7621	0.7613	0.7607	0.3297
0.8191	13.45	390	0.6808	0.7348	0.7679	0.7372	0.7150	0.3297
0.8468	13.79	400	0.6843	0.7670	0.7789	0.7677	0.7621	0.3297
0.8277	14.14	410	0.6630	0.7599	0.7660	0.7607	0.7578	0.3297
0.8159	14.48	420	0.6621	0.7599	0.7650	0.7608	0.7584	0.3297
0.8803	14.83	430	0.6731	0.7634	0.7683	0.7644	0.7623	0.3297

Framework versions

• PEFT 0.9.0
• Transformers 4.39.0.dev0
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2