whisper-large-kurmanji-v1

This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4280
• Wer: 10.8064

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: 16
• eval_batch_size: 16
• 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: 1000
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3459	0.4941	500	0.4129	23.7122
0.2412	0.9881	1000	0.3338	18.6864
0.1751	1.4822	1500	0.2937	15.9180
0.1799	1.9763	2000	0.2648	14.3789
0.1063	2.4704	2500	0.2722	13.9077
0.1035	2.9644	3000	0.2520	13.0673
0.0616	3.4585	3500	0.2739	12.4015
0.0686	3.9526	4000	0.2686	12.6026
0.0371	4.4466	4500	0.2964	12.7476
0.038	4.9407	5000	0.2906	12.5960
0.0182	5.4348	5500	0.3137	12.0324
0.0224	5.9289	6000	0.3099	12.7740
0.0083	6.4229	6500	0.3286	11.6930
0.0115	6.9170	7000	0.3322	11.8578
0.007	7.4111	7500	0.3465	12.2038
0.0087	7.9051	8000	0.3464	11.7259
0.0035	8.3992	8500	0.3756	11.4788
0.0053	8.8933	9000	0.3535	11.3832
0.0033	9.3874	9500	0.3805	11.2085
0.0037	9.8814	10000	0.3831	11.2645
0.0017	10.3755	10500	0.3754	11.1822
0.0024	10.8696	11000	0.3883	11.0899
0.001	11.3636	11500	0.3855	11.0075
0.0008	11.8577	12000	0.3890	11.2052
0.0004	12.3518	12500	0.3960	11.0338
0.0008	12.8458	13000	0.4007	10.7438
0.0001	13.3399	13500	0.4193	10.8526
0.0002	13.8340	14000	0.4219	10.7801
0.0001	14.3281	14500	0.4246	10.7636
0.0001	14.8221	15000	0.4280	10.8064

Framework versions

• Transformers 4.51.3
• Pytorch 2.4.1+cu124
• Datasets 3.6.0
• Tokenizers 0.21.2