steja/whisper-small-kannada

whisper-small-tamil

This model is a fine-tuned version of openai/whisper-small on the google/fleurs dataset for Kannada. It achieves the following results on the evaluation set:

• Loss: 0.2507
• Wer: 23.1257

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0792	2.27	500	0.2674	24.7048
0.0067	12.19	1000	0.1930	23.7758
0.0011	18.29	1500	0.2161	23.3225
0.0002	24.39	2000	0.2294	23.1332
0.0001	30.48	2500	0.2406	23.1652
0.0001	36.58	3000	0.2461	23.1531
0.0001	42.68	3500	0.2493	23.1108
0.0001	48.78	4000	0.2507	23.1257

Framework versions

• Transformers 4.24.0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1
• Tokenizers 0.13.2