hubert-large-ls960-ft-V2-50

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: 0.9222
• Wer: 0.0732
• Per: 0.0540

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
12.8849	1.0	818	4.5925	0.9551	0.9649
2.7511	2.0	1636	1.7073	0.4693	0.4560
1.1653	3.0	2454	1.1204	0.1534	0.1317
0.7529	4.0	3272	1.0336	0.1055	0.0841
0.6309	5.0	4090	1.0015	0.1023	0.0817
0.5354	6.0	4908	1.0387	0.0992	0.0777
0.4907	7.0	5726	0.9957	0.1087	0.0893
0.4326	8.0	6544	0.8882	0.1091	0.0844
0.4148	9.0	7362	0.9542	0.0830	0.0638
0.3779	10.0	8180	0.9479	0.0690	0.0501
0.3502	11.0	8998	0.9840	0.0689	0.0491
0.3294	12.0	9816	1.0877	0.0694	0.0491
0.3239	13.0	10634	0.8955	0.0731	0.0534
0.3069	14.0	11452	0.8547	0.0776	0.0580
0.2689	15.0	12270	0.9683	0.0720	0.0525
0.2486	16.0	13088	0.9282	0.0704	0.0519
0.2291	17.0	13906	0.9004	0.0671	0.0481
0.2294	18.0	14724	0.9242	0.0747	0.0547
0.2151	19.0	15542	0.9400	0.0747	0.0554
0.2109	20.0	16360	0.9222	0.0732	0.0540

Framework versions

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