tibetan-CS-detector_mbert-tibetan-continual-wylie_all_data_no_labels

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: 49.1782
• Accuracy: 0.9290
• Switch Precision: 0.4851
• Switch Recall: 0.8909
• Switch F1: 0.6282
• True Switches: 165
• Pred Switches: 303
• Exact Matches: 131
• Proximity Matches: 16
• To Auto Precision: 0.6050
• To Auto Recall: 0.9
• To Allo Precision: 0.4076
• To Allo Recall: 0.8824
• True To Auto: 80
• True To Allo: 85
• Matched To Auto: 72
• Matched To Allo: 75

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
13.5452	1.5789	30	3.9859	0.6703	0.1875	0.0182	0.0331	165	16	0	3	0.0769	0.0125	0.6667	0.0235	80	85	1	2
4.1536	3.1579	60	2.9644	0.7819	0.6667	0.0242	0.0468	165	6	4	0	0.8	0.05	0.0	0.0	80	85	4	0
7.9073	4.7368	90	3.2495	0.7947	0.7361	0.3212	0.4473	165	72	51	2	0.7391	0.6375	0.6667	0.0235	80	85	51	2
8.2183	6.3158	120	3.6442	0.7945	0.5175	0.4485	0.4805	165	143	68	6	0.6392	0.775	0.2609	0.1412	80	85	62	12
4.9603	7.8947	150	3.6908	0.7961	0.4466	0.5576	0.4960	165	206	84	8	0.6562	0.7875	0.2636	0.3412	80	85	63	29
8.0485	9.4737	180	2.1089	0.8634	0.6429	0.5455	0.5902	165	140	87	3	0.6667	0.85	0.5789	0.2588	80	85	68	22
2.8204	11.0526	210	4.9959	0.8964	0.4345	0.8242	0.5690	165	313	115	21	0.5854	0.9	0.3368	0.7529	80	85	72	64
5.5281	12.6316	240	4.1823	0.9059	0.4187	0.8424	0.5594	165	332	118	21	0.5373	0.9	0.3384	0.7882	80	85	72	67
5.8014	14.2105	270	4.7370	0.9123	0.4316	0.8606	0.5749	165	329	126	16	0.5414	0.9	0.3571	0.8235	80	85	72	70
2.0906	15.7895	300	17.8083	0.9123	0.4540	0.8667	0.5958	165	315	127	16	0.5902	0.9	0.3679	0.8353	80	85	72	71
1.9274	17.3684	330	22.6264	0.9191	0.4337	0.8727	0.5795	165	332	122	22	0.6	0.9	0.3396	0.8471	80	85	72	72
1.6201	18.9474	360	50.9304	0.9172	0.4398	0.8848	0.5875	165	332	129	17	0.5806	0.9	0.3558	0.8706	80	85	72	74
6.5566	20.5263	390	35.8194	0.9231	0.4660	0.8727	0.6076	165	309	130	14	0.5414	0.9	0.4091	0.8471	80	85	72	72
5.3539	22.1053	420	49.6696	0.9239	0.4492	0.8848	0.5959	165	325	131	15	0.5854	0.9	0.3663	0.8706	80	85	72	74
4.5666	23.6842	450	51.3558	0.9242	0.4635	0.8848	0.6083	165	315	133	13	0.5669	0.9	0.3936	0.8706	80	85	72	74
3.8105	25.2632	480	50.8006	0.9263	0.4273	0.8909	0.5776	165	344	129	18	0.5455	0.9	0.3538	0.8824	80	85	72	75
1.371	26.8421	510	48.9676	0.9266	0.4647	0.8788	0.6080	165	312	131	14	0.6050	0.9	0.3782	0.8588	80	85	72	73
9.9923	28.4211	540	73.3558	0.9265	0.4451	0.8848	0.5923	165	328	129	17	0.576	0.9	0.3645	0.8706	80	85	72	74
3.5917	30.0	570	58.4935	0.9270	0.4531	0.8788	0.5979	165	320	126	19	0.6050	0.9	0.3632	0.8588	80	85	72	73
7.5225	31.5789	600	72.4031	0.9273	0.4662	0.8788	0.6092	165	311	129	16	0.6	0.9	0.3822	0.8588	80	85	72	73
2.9639	33.1579	630	34.2685	0.9293	0.4851	0.8909	0.6282	165	303	131	16	0.6102	0.9	0.4054	0.8824	80	85	72	75
5.3705	34.7368	660	49.1782	0.9290	0.4851	0.8909	0.6282	165	303	131	16	0.6050	0.9	0.4076	0.8824	80	85	72	75

Framework versions

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