whisper-tiny-mn-9

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

• Loss: 0.3885
• Wer: 45.5102
• Cer: 17.3377

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: 64
• eval_batch_size: 64
• 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: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.587	0.69	1000	0.6937	75.6336	29.6764
0.4536	1.39	2000	0.5539	64.8187	24.8324
0.3798	2.08	3000	0.4963	57.7944	22.1842
0.3423	2.77	4000	0.4661	54.3751	20.9705
0.3122	3.47	5000	0.4449	52.5945	20.3405
0.3002	4.16	6000	0.4285	50.5080	19.3499
0.2842	4.85	7000	0.4171	49.3937	19.0282
0.2655	5.54	8000	0.4099	48.6727	18.6045
0.2555	6.24	9000	0.4035	48.2084	18.3392
0.2525	6.93	10000	0.3990	47.3290	17.8338
0.243	7.62	11000	0.3963	47.0559	18.2524
0.2358	8.32	12000	0.3948	46.7337	17.8186
0.2288	9.01	13000	0.3901	46.5480	17.9172
0.2171	9.7	14000	0.3910	46.0236	17.6266
0.2184	10.4	15000	0.3904	46.4387	17.8228
0.2099	11.09	16000	0.3893	45.9744	17.4379
0.216	11.78	17000	0.3889	45.6194	17.2939
0.2095	12.47	18000	0.3895	45.7887	17.4438
0.2056	13.17	19000	0.3882	45.6085	17.2888
0.2064	13.86	20000	0.3885	45.5102	17.3377

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2