whisper-medium-nyagen-balanced-62

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

• Loss: 0.4667
• Wer: 0.4238

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: 2
• eval_batch_size: 2
• seed: 62
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• 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: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4917	0.9864	200	0.6422	0.4577
0.2784	1.9716	400	0.4886	0.3640
0.1586	2.9568	600	0.4736	0.2924
0.0883	3.9420	800	0.4825	0.3126
0.037	4.9273	1000	0.4667	0.4238
0.0251	5.9125	1200	0.4977	0.2790
0.0116	6.8977	1400	0.5010	0.2717
0.0085	7.8829	1600	0.4902	0.2669
0.0058	8.8681	1800	0.5230	0.2597

Framework versions

• Transformers 4.53.0.dev0
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.0