BlueRaccoon/whisper-small-kab

Whisper Small Georgian

This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 kab dataset. It achieves the following results on the evaluation set:

• Loss: 0.5547
• Wer: 41.4753

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: 64
• eval_batch_size: 32
• seed: 42
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 400
• training_steps: 5000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4105	1.42	1000	0.5508	48.1244
0.258	2.84	2000	0.4905	42.0295
0.0998	4.27	3000	0.5220	41.6808
0.064	5.69	4000	0.5547	41.4753
0.0356	7.11	5000	0.5889	41.5737

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.1+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2