hubert-large-ll60k_arabic

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

• Loss: 4.0197
• Wer: 1.0
• Per: 1.0

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
15.0257	1.0	1637	4.0741	1.0	1.0
4.0101	2.0	3274	4.0442	1.0	1.0
3.9446	3.0	4911	4.0444	1.0	1.0
3.9302	4.0	6548	4.0440	1.0	1.0
3.9456	5.0	8185	4.0453	1.0	1.0
4.0388	6.0	9822	4.0415	1.0	1.0
3.9297	7.0	11459	4.0442	1.0	1.0
3.9443	8.0	13096	4.0395	1.0	1.0
3.9362	9.0	14733	4.0647	1.0000	1.0000
3.9785	10.0	16370	4.3751	1.0	1.0
3.9934	11.0	18007	4.0295	1.0	1.0
3.9361	12.0	19644	4.0276	1.0	1.0
3.913	13.0	21281	4.0239	1.0	1.0
3.9023	14.0	22918	4.0227	1.0	1.0
3.8957	15.0	24555	4.0228	1.0	1.0
3.9021	16.0	26192	4.0207	1.0	1.0
3.8897	17.0	27829	4.0213	1.0	1.0
3.8939	18.0	29466	4.0213	1.0	1.0
3.8949	19.0	31103	4.0204	1.0	1.0
3.8987	20.0	32740	4.0197	1.0	1.0

Framework versions

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