tibetan-CS-detector_mbert-tibetan-continual-wylie

This model is a fine-tuned version of OMRIDRORI/mbert-tibetan-continual-wylie-final on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 24.0864
• Accuracy: 0.9401
• Switch Precision: 0.4885
• Switch Recall: 0.9275
• Switch F1: 0.6400
• True Switches: 138
• Pred Switches: 262
• Exact Matches: 118
• Proximity Matches: 10
• To Auto Precision: 0.6238
• To Auto Recall: 0.9403
• To Allo Precision: 0.4037
• To Allo Recall: 0.9155
• True To Auto: 67
• True To Allo: 71
• Matched To Auto: 63
• Matched To Allo: 65

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: 1e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• num_epochs: 35
• mixed_precision_training: Native AMP
• label_smoothing_factor: 0.05

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Switch Precision	Switch Recall	Switch F1	True Switches	Pred Switches	Exact Matches	Proximity Matches	To Auto Precision	To Auto Recall	To Allo Precision	To Allo Recall	True To Auto	True To Allo	Matched To Auto	Matched To Allo
9.6194	1.9355	30	3.7660	0.6124	0.0	0.0	0.0	138	8	0	0	0.0	0.0	0.0	0.0	67	71	0	0
4.2982	3.8710	60	3.3089	0.7536	0.0833	0.0072	0.0133	138	12	1	0	0.2	0.0149	0.0	0.0	67	71	1	0
6.6871	5.8065	90	4.7429	0.7673	0.5543	0.3696	0.4435	138	92	51	0	0.6456	0.7612	0.0	0.0	67	71	51	0
5.4271	7.7419	120	4.9560	0.7681	0.4180	0.5725	0.4832	138	189	73	6	0.6522	0.8955	0.1959	0.2676	67	71	60	19
10.8753	9.6774	150	6.3715	0.7934	0.4081	0.6594	0.5042	138	223	84	7	0.6458	0.9254	0.2283	0.4085	67	71	62	29
3.1874	11.6129	180	5.5777	0.8773	0.5251	0.6812	0.5931	138	179	87	7	0.6392	0.9254	0.3902	0.4507	67	71	62	32
15.9945	13.5484	210	3.6266	0.9099	0.5	0.7971	0.6145	138	220	102	8	0.6562	0.9403	0.3790	0.6620	67	71	63	47
4.8693	15.4839	240	2.1193	0.9206	0.5497	0.7609	0.6383	138	191	94	11	0.62	0.9254	0.4725	0.6056	67	71	62	43
2.1608	17.4194	270	3.1227	0.9220	0.5081	0.9130	0.6528	138	248	114	12	0.63	0.9403	0.4257	0.8873	67	71	63	63
1.8562	19.3548	300	3.2563	0.9277	0.4330	0.9130	0.5874	138	291	106	20	0.5526	0.9403	0.3559	0.8873	67	71	63	63
1.8304	21.2903	330	27.3071	0.9305	0.4472	0.9203	0.6019	138	284	114	13	0.5625	0.9403	0.3721	0.9014	67	71	63	64
1.5658	23.2258	360	18.5545	0.9301	0.4498	0.9420	0.6089	138	289	120	10	0.5289	0.9552	0.3929	0.9296	67	71	64	66
1.5259	25.1613	390	2.7254	0.9352	0.4886	0.9348	0.6418	138	264	120	9	0.5943	0.9403	0.4177	0.9296	67	71	63	66
4.3114	27.0968	420	24.4862	0.9371	0.4813	0.9348	0.6355	138	268	120	9	0.6058	0.9403	0.4024	0.9296	67	71	63	66
2.6512	29.0323	450	24.4336	0.9375	0.4905	0.9348	0.6434	138	263	120	9	0.6364	0.9403	0.4024	0.9296	67	71	63	66
1.3451	30.9677	480	2.3665	0.9401	0.4868	0.9348	0.6402	138	265	120	9	0.5943	0.9403	0.4151	0.9296	67	71	63	66
1.2256	32.9032	510	24.0864	0.9401	0.4885	0.9275	0.6400	138	262	118	10	0.6238	0.9403	0.4037	0.9155	67	71	63	65

Framework versions

• Transformers 4.46.3
• Pytorch 2.4.1+cu121
• Datasets 2.0.0
• Tokenizers 0.20.3