SmartWhatt/rangsitra-whisper-sm-replace-0.5

Rangsitra Whisper

This model is a fine-tuned version of openai/whisper-small on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Cer: 6.8869
• Wer: 23.6206
• Loss: 0.2097

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
• 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: warmup_stable_decay
• lr_scheduler_warmup_steps: 1017
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Wer	Validation Loss
0.1667	1.0	2035	10.0135	33.9935	0.2194
0.1034	2.0	4070	9.2308	28.9993	0.1958
0.061	3.0	6105	8.1348	26.5631	0.2008
0.0205	4.0	8140	7.1528	24.4929	0.2052
0.0059	5.0	10175	6.8869	23.6206	0.2097

Framework versions

• Transformers 4.52.4
• Pytorch 2.9.1+cu128
• Datasets 3.6.0
• Tokenizers 0.21.4