hubert_arabic_mdd

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: 0.5363
• Wer: 0.0490
• Per: 0.0376

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
2.371	1.0	818	1.3106	0.6828	0.6863
0.9013	2.0	1636	0.4890	0.1144	0.0985
0.3369	3.0	2454	0.4295	0.0766	0.0611
0.2219	4.0	3272	0.4386	0.0649	0.0508
0.1565	5.0	4090	0.4124	0.0660	0.0506
0.1343	6.0	4908	0.4422	0.0630	0.0493
0.106	7.0	5726	0.4762	0.0600	0.0469
0.091	8.0	6544	0.4487	0.0580	0.0460
0.0745	9.0	7362	0.4284	0.0577	0.0461
0.0708	10.0	8180	0.4161	0.0578	0.0451
0.0621	11.0	8998	0.4659	0.0535	0.0414
0.0492	12.0	9816	0.5249	0.0557	0.0433
0.0479	13.0	10634	0.5411	0.0550	0.0426
0.0452	14.0	11452	0.5161	0.0536	0.0410
0.0385	15.0	12270	0.5002	0.0521	0.0404
0.0354	16.0	13088	0.4800	0.0499	0.0389
0.0342	17.0	13906	0.5079	0.0506	0.0394
0.0269	18.0	14724	0.5144	0.0499	0.0386
0.0247	19.0	15542	0.5334	0.0496	0.0380
0.0208	20.0	16360	0.5363	0.0490	0.0376

Framework versions

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