lorenzoncina/whisper-medium-ar

Whisper Medium Arabic

This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_11_0 ar dataset. It achieves the following results on the evaluation set:

• Loss: 0.4218
• Wer: 47.5307

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 4
• seed: 42
• 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
0.2215	0.1	1000	0.3361	49.9307
0.1134	1.07	2000	0.3290	56.76
0.0765	2.04	3000	0.3400	54.3947
0.0417	3.01	4000	0.3599	52.5320
0.0364	3.11	5000	0.3740	55.5653
0.0094	4.08	6000	0.4152	56.4307
0.0077	5.05	7000	0.4218	47.5307
0.0018	6.02	8000	0.4556	50.0493
0.0012	6.12	9000	0.4760	54.8147
0.0009	7.09	10000	0.4711	48.7533

Framework versions

• Transformers 4.28.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.11.1.dev0
• Tokenizers 0.13.2