neddamj/whisper-base-patois

Whisper base - Patois

This model is a fine-tuned version of openai/whisper-base on the Patois Music Transcription dataset. It achieves the following results on the evaluation set:

• Loss: 1.0740
• Wer: 0.5879

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: 8
• seed: 42
• optimizer: Use OptimizerNames.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: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.3852	1.3746	400	1.3994	0.8946
0.9297	2.7491	800	1.1046	0.7298
0.5738	4.1237	1200	1.0175	0.6647
0.4621	5.4983	1600	0.9841	0.6448
0.3866	6.8729	2000	0.9733	0.5982
0.2647	8.2474	2400	0.9925	0.6296
0.2166	9.6220	2800	1.0156	0.6006
0.185	10.9966	3200	1.0366	0.5932
0.1521	12.3711	3600	1.0650	0.5983
0.1168	13.7457	4000	1.0740	0.5879

Framework versions

• Transformers 4.52.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1