whisper-small-serbian-v2

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2957
• Wer: 14.0126

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: 3e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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: 1250
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.304	0.9234	500	0.2736	21.4309
0.2213	1.8458	1000	0.2192	17.9671
0.1528	2.7682	1500	0.2149	18.1085
0.0876	3.6907	2000	0.2211	16.4412
0.0598	4.6131	2500	0.2254	15.9645
0.0335	5.5355	3000	0.2396	15.7533
0.0172	6.4580	3500	0.2494	15.8825
0.0109	7.3804	4000	0.2568	15.4303
0.0067	8.3029	4500	0.2684	15.4198
0.0038	9.2253	5000	0.2710	14.9571
0.0036	10.1477	5500	0.2738	14.9327
0.0019	11.0702	6000	0.2772	14.7406
0.0019	11.9935	6500	0.2744	14.8262
0.0018	12.9160	7000	0.2770	14.5102
0.0011	13.8384	7500	0.2784	14.3688
0.0005	14.7608	8000	0.2814	14.2762
0.0002	15.6833	8500	0.2870	14.1767
0.0002	16.6057	9000	0.2904	14.1523
0.0001	17.5282	9500	0.2934	14.0563
0.0001	18.4506	10000	0.2944	13.9742
0.0001	19.3730	10500	0.2957	14.0126

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.2