whisper-medium-swagen-balanced-62

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

• Loss: 0.5541
• Wer: 0.3919

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
0.6238	0.4770	200	0.8122	0.4737
0.4914	0.9541	400	0.6395	0.4002
0.3098	1.4293	600	0.5996	0.3495
0.2721	1.9064	800	0.5564	0.3403
0.1282	2.3816	1000	0.5693	0.3892
0.1367	2.8587	1200	0.5541	0.3919
0.0388	3.3339	1400	0.5811	0.3095
0.0526	3.8110	1600	0.5600	0.3111
0.0193	4.2862	1800	0.5913	0.2999
0.0251	4.7633	2000	0.5872	0.3650

Framework versions

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