hubert_arabic_mdd_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: 0.5409
• Wer: 0.0470
• Per: 0.0360

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
8.874	1.0	818	2.7367	1.0	1.0
1.4558	2.0	1636	0.6219	0.2525	0.2344
0.3775	3.0	2454	0.4014	0.0885	0.0742
0.2126	4.0	3272	0.4332	0.0656	0.0500
0.1658	5.0	4090	0.4970	0.0608	0.0464
0.1308	6.0	4908	0.4520	0.0591	0.0445
0.1058	7.0	5726	0.4513	0.0531	0.0407
0.0968	8.0	6544	0.4253	0.0548	0.0413
0.0871	9.0	7362	0.4232	0.0524	0.0405
0.0681	10.0	8180	0.4612	0.0508	0.0392
0.0651	11.0	8998	0.4978	0.0516	0.0394
0.0587	12.0	9816	0.5069	0.0512	0.0389
0.046	13.0	10634	0.5170	0.0489	0.0370
0.0439	14.0	11452	0.5758	0.0536	0.0417
0.0401	15.0	12270	0.5238	0.0500	0.0384
0.0346	16.0	13088	0.5668	0.0483	0.0368
0.0334	17.0	13906	0.5343	0.0459	0.0350
0.0336	18.0	14724	0.5242	0.0466	0.0361
0.0266	19.0	15542	0.5379	0.0467	0.0359
0.0245	20.0	16360	0.5409	0.0470	0.0360

Framework versions

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