whisper-medium-swagen-combined-15hrs-model

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: 0.4103
• Wer: 0.2717

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: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• 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
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.6268	0.1654	200	0.8031	0.4605
2.0712	0.3308	400	0.6148	0.3829
1.7302	0.4962	600	0.5562	0.3490
1.5735	0.6616	800	0.5103	0.3106
1.5623	0.8270	1000	0.4683	0.2776
1.2713	0.9924	1200	0.4439	0.2688
0.7209	1.1571	1400	0.4601	0.2732
0.6856	1.3225	1600	0.4391	0.2595
0.7661	1.4879	1800	0.4396	0.2755
0.8113	1.6533	2000	0.4262	0.2643
0.77	1.8187	2200	0.4175	0.2679
0.6942	1.9841	2400	0.4103	0.2717
0.2814	2.1489	2600	0.4295	0.2617
0.3171	2.3142	2800	0.4301	0.2432
0.3495	2.4796	3000	0.4299	0.2526

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0