hubert-xlarge-ll60k_arabic

This model is a fine-tuned version of facebook/hubert-xlarge-ll60k on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.1133
• Wer: 0.6646
• Per: 0.6706

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
13.7253	1.0	1637	3.3328	1.0	1.0
3.3354	2.0	3274	3.2847	1.0	1.0
3.304	3.0	4911	3.2375	1.0	1.0
3.2655	4.0	6548	3.2143	1.0	1.0
3.2242	5.0	8185	3.1874	1.0	1.0
3.1556	6.0	9822	3.0734	1.0	1.0
3.0485	7.0	11459	2.9548	1.0	1.0
2.935	8.0	13096	2.8378	0.9043	0.9171
2.8158	9.0	14733	2.7023	0.8971	0.9087
2.7185	10.0	16370	2.5982	0.8865	0.8997
2.6406	11.0	18007	2.5032	0.8453	0.8571
2.5721	12.0	19644	2.4385	0.8179	0.8291
2.5136	13.0	21281	2.3810	0.7994	0.8097
2.457	14.0	22918	2.3041	0.7792	0.7896
2.4104	15.0	24555	2.2466	0.7588	0.7695
2.3711	16.0	26192	2.2066	0.7321	0.7417
2.3365	17.0	27829	2.1757	0.7002	0.7072
2.3109	18.0	29466	2.1412	0.6771	0.6828
2.2857	19.0	31103	2.1265	0.6722	0.6785
2.2757	20.0	32740	2.1133	0.6646	0.6706

Framework versions

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