adithyal1998Bhat/whisper-kn

Whisper Small kn - Saraswathi

This model is a fine-tuned version of ope100whisper-small on the kannada voices dataset. It achieves the following results on the evaluation set:

• Loss: 0.1305
• Wer: 24.4986

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: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• 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
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1461	0.5869	1000	0.1511	37.9110
0.0795	1.1737	2000	0.1172	31.0520
0.0715	1.7613	3000	0.1090	28.1220
0.0508	2.3486	4000	0.1033	25.7362
0.0309	2.9356	5000	0.1101	25.1920
0.0474	3.5230	6000	0.1105	26.1537
0.0272	4.1098	7000	0.1169	25.4082
0.0255	4.6967	8000	0.1195	25.0727
0.0151	5.2835	9000	0.1285	24.7968
0.0149	5.8704	10000	0.1305	24.4986

Framework versions

• Transformers 4.48.0
• Pytorch 2.7.0+cu126
• Datasets 3.6.0
• Tokenizers 0.21.2