whisper-kazakh-output

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

• Loss: 0.2160
• Wer: 0.286

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: 0.0005
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 6
• total_train_batch_size: 192
• total_eval_batch_size: 32
• optimizer: Use adamw_torch_fused 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: 55
• training_steps: 663

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
21.0919	0.1659	110	0.3962	1.898
14.5363	0.3318	220	0.2967	0.4237
10.9490	0.4977	330	0.2511	0.3364
8.2949	0.6637	440	0.2335	0.3201
6.7211	0.8296	550	0.2190	0.2833
5.8075	0.9955	660	0.2160	0.286

Framework versions

• PEFT 0.18.1
• Transformers 5.3.0
• Pytorch 2.10.0+cu128
• Datasets 4.6.1
• Tokenizers 0.22.2