hubert-large-ls960-ft

This model is a fine-tuned version of facebook/hubert-large-ls960-ft on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.4850
• Wer: 0.7884
• Per: 0.7945

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
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Per
14.5096	1.0	1637	3.3563	1.0	1.0
3.3754	2.0	3274	3.3135	1.0	1.0
3.3037	3.0	4911	3.2654	1.0	1.0
3.2641	4.0	6548	3.2260	1.0	1.0
3.1723	5.0	8185	3.1158	0.9975	0.9980
3.0864	6.0	9822	3.0251	0.9105	0.9234
3.0205	7.0	11459	2.9357	0.8957	0.9068
2.9474	8.0	13096	2.8523	0.8456	0.8528
2.883	9.0	14733	2.7752	0.8260	0.8323
2.8246	10.0	16370	2.7173	0.8138	0.8202
2.7959	11.0	18007	2.6877	0.8074	0.8139
2.7595	12.0	19644	2.6459	0.8131	0.8224
2.7335	13.0	21281	2.6208	0.8050	0.8120
2.7014	14.0	22918	2.5800	0.7986	0.8054
2.6792	15.0	24555	2.5604	0.7989	0.8058
2.6539	16.0	26192	2.5289	0.7929	0.7985
2.6316	17.0	27829	2.5119	0.7943	0.8018
2.6212	18.0	29466	2.4965	0.7931	0.8005
2.6114	19.0	31103	2.4890	0.7894	0.7959
2.6028	20.0	32740	2.4850	0.7884	0.7945

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0