whisper-large-v2-ru-tuned

This model was trained from scratch on the common_voice_11_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1714
• Wer Ortho: 13.2854
• Wer: 9.7737

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
• 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: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 12000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Wer Ortho
0.0806	0.1231	500	0.1944	10.7828	15.1977
0.1386	0.2462	1000	0.1729	9.8746	13.9468
0.1309	0.3693	1500	0.1624	10.0187	14.2056
0.1101	0.4924	2000	0.1543	9.7160	13.4292
0.1589	0.6155	2500	0.1571	9.6007	13.4148
0.1479	0.7386	3000	0.1553	13.6017	9.9611
0.1133	0.8616	3500	0.1535	12.6671	9.4133
0.1472	0.9847	4000	0.1472	12.3940	9.0529
0.0584	1.1078	4500	0.1567	12.9260	9.2547
0.064	1.2309	5000	0.1569	13.7168	9.9178
0.0657	1.3540	5500	0.1713	14.0187	10.4368
0.0712	1.4771	6000	0.1664	14.4069	10.7539
0.0793	1.6002	6500	0.1607	12.7678	9.1106
0.068	1.7233	7000	0.1637	12.3364	8.8078
0.0646	1.8464	7500	0.1623	13.0122	9.5286
0.0747	1.9695	8000	0.1580	12.3652	9.2691
0.0346	2.0926	8500	0.1674	12.7534	9.4133
0.04	2.2157	9000	0.1725	13.2135	9.1826
0.0346	2.3387	9500	0.1710	12.6096	8.8655
0.0418	2.4618	10000	0.1771	14.6226	10.8837
0.0482	2.5849	10500	0.1688	13.2135	9.5863
0.0464	2.7080	11000	0.1774	13.6592	9.9899
0.0358	2.8311	11500	0.1731	13.0841	9.4710
0.044	2.9542	12000	0.1714	13.2854	9.7737

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0