whisper-synthesized-turkish-4-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: 0.4388
• Wer: 15.4240

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: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7343	1.04	100	0.2580	16.7757
0.1568	2.08	200	0.2714	15.3129
0.091	3.12	300	0.3099	16.2573
0.0861	4.17	400	0.3946	22.9910
0.0967	5.21	500	0.4884	23.2132
0.0775	6.25	600	0.4263	19.9852
0.0692	7.29	700	0.4428	20.0099
0.052	8.33	800	0.4407	23.6761
0.0458	9.38	900	0.4760	19.7445
0.0326	10.42	1000	0.4847	18.6520
0.0281	11.46	1100	0.4936	20.2074
0.0221	12.5	1200	0.4655	19.3495
0.0123	13.54	1300	0.4657	17.5781
0.0105	14.58	1400	0.4493	16.2264
0.0042	15.62	1500	0.4396	15.5660
0.0029	16.67	1600	0.4412	15.7882
0.0011	17.71	1700	0.4400	15.8190
0.0005	18.75	1800	0.4400	15.4672
0.0003	19.79	1900	0.4389	15.4117
0.0002	20.83	2000	0.4388	15.4240

Framework versions

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