npallewela/whisper-small-ap3

Whisper small ap3 - Nuwan

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5516
• Wer Ortho: 31.1588
• Wer: 30.4553

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-06
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.4586	0.1642	400	0.5965	36.5712	35.9906
0.45	0.3284	800	0.5949	34.1257	33.5673
0.4665	0.4926	1200	0.5824	33.6163	33.0131
0.4427	0.6568	1600	0.5769	34.0227	33.3670
0.3917	0.8210	2000	0.5714	32.6097	31.9060
0.423	0.9852	2400	0.5660	32.9462	32.3243
0.3857	1.1494	2800	0.5634	31.4802	30.8823
0.372	1.3136	3200	0.5638	31.6394	30.9379
0.3803	1.4778	3600	0.5603	31.0617	30.4553
0.356	1.6420	4000	0.5516	31.1588	30.4553

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.0+cu126
• Datasets 3.6.0
• Tokenizers 0.22.1