Marialab/whisper-small-dr-ar-mish

Whisper small darija translate

This model is a fine-tuned version of openai/whisper-small on the Darija-C dataset. It achieves the following results on the evaluation set:

• Loss: 0.0001
• Bleu: 0.0681

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: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 4
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2000
• training_steps: 16000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
0.6324	133.3333	2000	0.5772	0.0
0.2913	266.6667	4000	0.2546	0.1229
0.2803	400.0	6000	0.2447	0.0859
0.2342	533.3333	8000	0.2505	0.0428
0.0562	666.6667	10000	0.0231	0.1789
0.0002	800.0	12000	0.0002	0.1136
0.0001	933.3333	14000	0.0001	0.0681
0.0001	1066.6667	16000	0.0001	0.0681

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1+cu121
• Datasets 2.19.2
• Tokenizers 0.20.3