lutskivd/whisper-russian-finetune-v2

Whisper tiny Russian v2

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

• Loss: 0.2382
• Wer: 9.6618

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: 2
• eval_batch_size: 1
• 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: 1024
• training_steps: 4000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3414	0.5353	500	0.3009	16.2980
0.2112	1.0707	1000	0.2510	11.4416
0.1769	1.6060	1500	0.2422	10.3992
0.0752	2.1413	2000	0.2349	10.2721
0.0456	2.6767	2500	0.2336	10.1704
0.0215	3.2120	3000	0.2368	9.8398
0.0169	3.7473	3500	0.2354	9.7381
0.0035	4.2827	4000	0.2382	9.6618

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1