whisper-synthesized-turkish-8-hour-hlr

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: 0.3824
• Wer: 49.2902

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: 0.0001
• 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: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7481	0.52	100	0.2675	14.6834
0.1975	1.04	200	0.2534	13.2144
0.1388	1.56	300	0.2755	15.6647
0.1585	2.08	400	0.3080	14.6649
0.1153	2.6	500	0.3421	17.7447
0.1241	3.12	600	0.3570	16.8189
0.1093	3.65	700	0.3776	18.8125
0.09	4.17	800	0.3859	30.0518
0.0751	4.69	900	0.3874	17.3929
0.0758	5.21	1000	0.3987	20.0901
0.0602	5.73	1100	0.4017	17.1460
0.0568	6.25	1200	0.3824	15.6154
0.0454	6.77	1300	0.3926	15.8808
0.0433	7.29	1400	0.4146	16.3869
0.0341	7.81	1500	0.4078	16.1153
0.0295	8.33	1600	0.4192	17.1275
0.0274	8.85	1700	0.4140	16.3745
0.0246	9.38	1800	0.4077	21.0344
0.0211	9.9	1900	0.4003	19.8741
0.0149	10.42	2000	0.4054	108.7335
0.0172	10.94	2100	0.3917	20.6024
0.0138	11.46	2200	0.3942	889.4643
0.0108	11.98	2300	0.3906	55.0673
0.0099	12.5	2400	0.3834	29.9778
0.0067	13.02	2500	0.3947	34.5883
0.0045	13.54	2600	0.3940	20.9789
0.0035	14.06	2700	0.3911	15.6462
0.0031	14.58	2800	0.3905	18.3990
0.0018	15.1	2900	0.3919	16.3190
0.0011	15.62	3000	0.3906	18.0286
0.001	16.15	3100	0.3911	17.6521
0.0006	16.67	3200	0.3813	27.6879
0.0007	17.19	3300	0.3800	45.7536
0.0003	17.71	3400	0.3805	51.2529
0.0001	18.23	3500	0.3815	51.7282
0.0001	18.75	3600	0.3821	47.0065
0.0002	19.27	3700	0.3821	45.8585
0.0001	19.79	3800	0.3823	47.7904
0.0001	20.31	3900	0.3824	49.2594
0.0003	20.83	4000	0.3824	49.2902

Framework versions

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