whisper-large-turkish-v6

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:

Model description

More information needed

More information needed

More information needed

The following hyperparameters were used during training:

learning_rate: 1e-05
train_batch_size: 16
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: linear
lr_scheduler_warmup_steps: 1500
num_epochs: 12
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer
0.1865	0.3680	1500	0.1917	16.9584
0.1707	0.7360	3000	0.1882	16.4130
0.1248	1.1040	4500	0.1851	15.9936
0.1096	1.4720	6000	0.1901	15.8933
0.1269	1.8400	7500	0.1803	15.3985
0.0753	2.2080	9000	0.1928	16.3164
0.0793	2.5761	10500	0.1905	15.4878
0.0817	2.9441	12000	0.1862	16.0507
0.0455	3.3121	13500	0.2044	15.4315
0.0493	3.6801	15000	0.2017	15.2111
0.0221	4.0481	16500	0.2220	15.3297
0.0245	4.4161	18000	0.2217	15.7953
0.0245	4.7841	19500	0.2284	15.2865
0.0115	5.1521	21000	0.2452	14.9110
0.0131	5.5201	22500	0.2391	14.9674
0.0137	5.8881	24000	0.2388	15.5083
0.0058	6.2561	25500	0.2544	14.8847
0.0063	6.6241	27000	0.2539	14.8627
0.0062	6.9921	28500	0.2661	15.2250
0.0047	7.3602	30000	0.2638	14.8086
0.0038	7.7282	31500	0.2751	14.9571
0.0018	8.0962	33000	0.2799	14.6197
0.0026	8.4642	34500	0.2776	14.8656
0.0014	8.8322	36000	0.2874	14.5150
0.0015	9.2002	37500	0.2841	14.5055
0.0022	9.5682	39000	0.2965	14.6570
0.0005	9.9362	40500	0.3031	14.4089
0.0003	10.3042	42000	0.3126	14.2486
0.0003	10.6722	43500	0.3153	14.2325
0.0002	11.0402	45000	0.3122	14.1622
0.0001	11.4082	46500	0.3201	14.1593
0.0001	11.7763	48000	0.3222	14.1454

Safetensors

Model size

2B params

Tensor type

F32

Base model

Finetuned

(763)

this model