whisper-synthesized-turkish-4-hour-llr

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: 0.2356
• Wer: 14.9179

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-06
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.8278	1.04	100	1.5357	41.2110
1.2048	2.08	200	0.8059	86.3165
0.6206	3.12	300	0.5751	94.1612
0.4927	4.17	400	0.5024	80.1568
0.4277	5.21	500	0.4451	48.6607
0.3502	6.25	600	0.3999	32.7490
0.2948	7.29	700	0.3650	21.5961
0.24	8.33	800	0.3344	17.7756
0.2079	9.38	900	0.3010	15.2512
0.1416	10.42	1000	0.2402	13.8687
0.1076	11.46	1100	0.2357	14.4427
0.0905	12.5	1200	0.2350	14.0415
0.0759	13.54	1300	0.2348	14.0230
0.0819	14.58	1400	0.2345	14.0600
0.0608	15.62	1500	0.2346	14.3563
0.0712	16.67	1600	0.2350	14.2575
0.0608	17.71	1700	0.2355	14.8562
0.0632	18.75	1800	0.2354	14.9241
0.0544	19.79	1900	0.2356	14.9303
0.0529	20.83	2000	0.2356	14.9179

Framework versions

• Transformers 4.28.0
• Pytorch 2.0.0+cu118
• Datasets 2.11.0
• Tokenizers 0.13.3