misterkissi/whisper-small-malagasy

wisper-small-malagasy

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

• Loss: 0.8951
• Wer: 0.4742

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: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.3904	8.1983	500	0.6457	0.4847
0.0285	16.3967	1000	0.7599	0.4859
0.0038	24.5950	1500	0.7853	0.4771
0.0008	32.7934	2000	0.8225	0.4738
0.0003	40.9917	2500	0.8441	0.4738
0.0002	49.1818	3000	0.8617	0.4754
0.0002	57.3802	3500	0.8751	0.4758
0.0002	65.5785	4000	0.8857	0.4746
0.0001	73.7769	4500	0.8922	0.4734
0.0001	81.9752	5000	0.8951	0.4742

Framework versions

• Transformers 4.53.2
• Pytorch 2.6.0+cu124
• Datasets 2.18.0
• Tokenizers 0.21.2