hubert-large-ls960-ft-V2-10

This model is a fine-tuned version of facebook/hubert-large-ls960-ft on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4727
• Wer: 0.0628
• Per: 0.0490

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: 0.0001
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Per
14.6873	1.0	164	3.2286	1.0	1.0
3.2356	2.0	328	3.1024	1.0	1.0
2.6803	3.0	492	1.9833	0.9224	0.9382
1.5141	4.0	656	1.1059	0.4099	0.3923
0.7862	5.0	820	0.6129	0.1587	0.1357
0.4004	6.0	984	0.4939	0.0939	0.0748
0.2791	7.0	1148	0.4888	0.0815	0.0646
0.2168	8.0	1312	0.5083	0.0830	0.0662
0.1726	9.0	1476	0.4748	0.0749	0.0596
0.1412	10.0	1640	0.4955	0.0742	0.0575
0.1156	11.0	1804	0.4986	0.0715	0.0564
0.1321	12.0	1968	0.5101	0.0703	0.0552
0.103	13.0	2132	0.4728	0.0668	0.0527
0.0772	14.0	2296	0.4832	0.0649	0.0509
0.0858	15.0	2460	0.4830	0.0649	0.0505
0.0874	16.0	2624	0.4697	0.0642	0.0509
0.0784	17.0	2788	0.4499	0.0652	0.0516
0.0703	18.0	2952	0.4699	0.0638	0.0500
0.062	19.0	3116	0.4757	0.0634	0.0496
0.0588	20.0	3280	0.4727	0.0628	0.0490

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0