arun100/whisper-base-ar-2

Whisper Base Arabic Derived

This model is a fine-tuned version of arun100/whisper-base-ar-1 on the google/fleurs ar_eg dataset. It achieves the following results on the evaluation set:

• Loss: 0.6670
• Wer: 44.2880

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: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5692	52.63	500	0.6253	54.7894
0.3447	105.26	1000	0.6001	45.2106
0.2067	157.89	1500	0.6109	44.7372
0.1273	210.53	2000	0.6303	44.7372
0.0788	263.16	2500	0.6508	44.4579
0.0526	315.79	3000	0.6670	44.2880
0.0404	368.42	3500	0.6784	44.7129
0.0335	421.05	4000	0.6860	46.2668
0.0296	473.68	4500	0.6907	44.5915
0.0287	526.32	5000	0.6924	44.6279

Framework versions

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