tibetan-CS-detector

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: 2.8365
• Accuracy: 0.9388
• Switch Precision: 0.4980
• Switch Recall: 0.9130
• Switch F1: 0.6445
• True Switches: 138
• Pred Switches: 253
• Exact Matches: 122
• Proximity Matches: 4
• To Auto Precision: 0.6966
• To Auto Recall: 0.9254
• To Allo Precision: 0.3902
• To Allo Recall: 0.9014
• True To Auto: 67
• True To Allo: 71
• Matched To Auto: 62
• Matched To Allo: 64

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
6.9424	1.9355	30	3.9697	0.4816	0.0	0.0	0.0	138	8	0	0	0.0	0.0	0.0	0.0	67	71	0	0
4.7989	3.8710	60	3.2594	0.7331	0.0	0.0	0.0	138	1	0	0	0.0	0.0	0.0	0.0	67	71	0	0
9.9599	5.8065	90	3.9145	0.7658	0.5909	0.2826	0.3824	138	66	39	0	0.6786	0.5672	0.1	0.0141	67	71	38	1
7.1635	7.7419	120	4.4059	0.7665	0.3818	0.4565	0.4158	138	165	62	1	0.6438	0.7015	0.1739	0.2254	67	71	47	16
10.5361	9.6774	150	5.7618	0.7737	0.3556	0.6159	0.4509	138	239	82	3	0.6667	0.8358	0.1871	0.4085	67	71	56	29
9.5003	11.6129	180	4.0246	0.8587	0.5741	0.4493	0.5041	138	108	62	0	0.7237	0.8209	0.2188	0.0986	67	71	55	7
11.3652	13.5484	210	3.3524	0.9056	0.4911	0.6014	0.5407	138	169	82	1	0.6818	0.8955	0.2840	0.3239	67	71	60	23
4.7329	15.4839	240	2.6446	0.9111	0.5337	0.6304	0.5781	138	163	85	2	0.6667	0.8955	0.3699	0.3803	67	71	60	27
2.2142	17.4194	270	4.7999	0.9163	0.5	0.8406	0.6270	138	232	114	2	0.6778	0.9104	0.3873	0.7746	67	71	61	55
6.1957	19.3548	300	2.5471	0.9232	0.5928	0.8333	0.6928	138	194	113	2	0.6932	0.9104	0.5094	0.7606	67	71	61	54
6.6179	21.2903	330	2.7181	0.9266	0.5619	0.8551	0.6782	138	210	116	2	0.6977	0.8955	0.4677	0.8169	67	71	60	58
1.6293	23.2258	360	2.1611	0.9365	0.4939	0.8768	0.6319	138	245	118	3	0.6813	0.9254	0.3831	0.8310	67	71	62	59
1.7535	25.1613	390	2.1557	0.9381	0.5105	0.8841	0.6472	138	239	119	3	0.7093	0.9104	0.3987	0.8592	67	71	61	61
1.4616	27.0968	420	3.3062	0.9368	0.4808	0.9058	0.6281	138	260	121	4	0.6966	0.9254	0.3684	0.8873	67	71	62	63
10.5341	29.0323	450	2.8365	0.9388	0.4980	0.9130	0.6445	138	253	122	4	0.6966	0.9254	0.3902	0.9014	67	71	62	64

Framework versions

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