erjoy/whisper-tiny-hi-3k-steps

Whisper Small Hi - V1

This model is a fine-tuned version of openai/whisper-tiny on the ai4bharat/IndicVoices dataset. It achieves the following results on the evaluation set:

• Loss: 0.4275
• Wer: 51.3495

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: 8
• 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: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6607	0.8104	500	0.6622	72.1794
0.4574	1.6207	1000	0.5104	59.5822
0.3659	2.4311	1500	0.4626	56.2291
0.3346	3.2415	2000	0.4417	53.4954
0.2912	4.0519	2500	0.4301	51.6118
0.275	4.8622	3000	0.4275	51.3495

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.1
• Tokenizers 0.21.0