tibetan-CS-detector_mbert-tibetan-continual-wylie_all_data_tol_10w

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: 66.6655
• Accuracy: 0.9211
• Switch Precision: 0.3780
• Switch Recall: 0.9576
• Switch F1: 0.5420
• True Switches: 165
• Pred Switches: 418
• Exact Matches: 128
• Proximity Matches: 30
• To Auto Precision: 0.5441
• To Auto Recall: 0.925
• To Allo Precision: 0.2979
• To Allo Recall: 0.9882
• True To Auto: 80
• True To Allo: 85
• Matched To Auto: 74
• Matched To Allo: 84

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
7.0857	1.5789	30	3.6157	0.7480	0.0	0.0	0.0	165	2	0	0	0.0	0.0	0.0	0.0	80	85	0	0
4.1349	3.1579	60	3.0511	0.7898	0.0	0.0	0.0	165	2	0	0	0.0	0.0	0.0	0.0	80	85	0	0
3.6816	4.7368	90	3.4370	0.7948	0.6383	0.3636	0.4633	165	94	58	2	0.7108	0.7375	0.0909	0.0118	80	85	59	1
8.0623	6.3158	120	10.2609	0.7935	0.4278	0.4848	0.4545	165	187	70	10	0.6505	0.8375	0.1548	0.1529	80	85	67	13
5.9021	7.8947	150	11.7316	0.7978	0.3664	0.6485	0.4683	165	292	89	18	0.6518	0.9125	0.1889	0.4	80	85	73	34
18.5865	9.4737	180	12.3717	0.8671	0.5694	0.4970	0.5307	165	144	74	8	0.6218	0.925	0.32	0.0941	80	85	74	8
10.0461	11.0526	210	20.2670	0.8958	0.5080	0.5758	0.5398	165	187	86	9	0.5968	0.925	0.3333	0.2471	80	85	74	21
2.3131	12.6316	240	22.5944	0.9021	0.4572	0.8424	0.5928	165	304	124	15	0.5103	0.925	0.4088	0.7647	80	85	74	65
5.9862	14.2105	270	42.4097	0.9130	0.4532	0.9091	0.6048	165	331	131	19	0.5362	0.925	0.3938	0.8941	80	85	74	76
7.8056	15.7895	300	13.1528	0.9111	0.4053	0.9333	0.5651	165	380	127	27	0.6116	0.925	0.3089	0.9412	80	85	74	80
4.8098	17.3684	330	39.5141	0.9141	0.4270	0.9212	0.5835	165	356	136	16	0.6167	0.925	0.3305	0.9176	80	85	74	78
1.6387	18.9474	360	54.7048	0.9201	0.4251	0.9455	0.5865	165	367	130	26	0.5522	0.925	0.3519	0.9647	80	85	74	82
10.5276	20.5263	390	54.1588	0.9245	0.4278	0.9515	0.5902	165	367	128	29	0.5606	0.925	0.3532	0.9765	80	85	74	83
1.4039	22.1053	420	66.6655	0.9211	0.3780	0.9576	0.5420	165	418	128	30	0.5441	0.925	0.2979	0.9882	80	85	74	84

Framework versions

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