arun100/whisper-base-ar-1

Whisper Base Arabic

This model is a fine-tuned version of openai/whisper-base on the mozilla-foundation/common_voice_16_0 ar dataset. It achieves the following results on the evaluation set:

• Loss: 0.5856
• Wer: 80.4777

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: 5e-07
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7392	1.53	500	0.8623	100.8133
0.5938	3.07	1000	0.7397	93.6651
0.5388	4.6	1500	0.6953	92.3005
0.4982	6.13	2000	0.6682	88.9392
0.4795	7.67	2500	0.6512	90.1524
0.4483	9.2	3000	0.6373	87.1234
0.4374	10.74	3500	0.6261	85.3144
0.4331	12.27	4000	0.6179	86.4290
0.4125	13.8	4500	0.6106	83.2865
0.3984	15.34	5000	0.6059	83.0676
0.4035	16.87	5500	0.6008	82.2165
0.3997	18.4	6000	0.5970	81.1195
0.3878	19.94	6500	0.5941	81.7153
0.3827	21.47	7000	0.5906	81.2559
0.3785	23.01	7500	0.5892	81.0506
0.372	24.54	8000	0.5882	81.4248
0.3655	26.07	8500	0.5865	81.0479
0.3697	27.61	9000	0.5856	80.4777
0.3658	29.14	9500	0.5849	80.6128
0.3539	30.67	10000	0.5848	80.6696

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0