sulaimank/whisper-small-lg-CVGRAIN-v4

whisper-small-lg-cv_grain_combined-v4

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: 0.0633
• Wer: 0.0537
• Cer: 0.0117

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: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use adamw_hf with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 80
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.4473	1.0	5827	0.3632	0.8771	0.2975
0.4358	2.0	11654	0.1614	0.6866	0.2496
0.2841	3.0	17481	0.0955	0.5376	0.2382
0.2019	4.0	23308	0.0713	0.4945	0.3163
0.1433	5.0	29135	0.0579	0.2610	0.1198
0.0996	6.0	34962	0.0536	0.1695	0.0514
0.0685	7.0	40789	0.0484	0.1309	0.0304
0.0495	8.0	46616	0.0461	0.0957	0.0224
0.0363	9.0	52443	0.0527	0.0835	0.0183
0.0256	10.0	58270	0.0535	0.0748	0.0177
0.0196	11.0	64097	0.0520	0.0798	0.0219
0.0158	12.0	69924	0.0527	0.0729	0.0171
0.0131	13.0	75751	0.0520	0.0686	0.0164
0.011	14.0	81578	0.0605	0.0630	0.0147
0.0098	15.0	87405	0.0533	0.0586	0.0136
0.0084	16.0	93232	0.0614	0.0630	0.0141
0.0076	17.0	99059	0.0642	0.0537	0.0133
0.0067	18.0	104886	0.0496	0.0566	0.0137
0.0062	19.0	110713	0.0597	0.0607	0.0144
0.0054	20.0	116540	0.0592	0.0580	0.0132
0.0049	21.0	122367	0.0447	0.0518	0.0127
0.0045	22.0	128194	0.0583	0.0501	0.0122
0.0041	23.0	134021	0.0667	0.0551	0.0120
0.0038	24.0	139848	0.0609	0.0534	0.0125
0.0036	25.0	145675	0.0539	0.0510	0.0121
0.0033	26.0	151502	0.0601	0.0506	0.0121
0.0031	27.0	157329	0.0567	0.0477	0.0125
0.0027	28.0	163156	0.0583	0.0518	0.0127
0.0025	29.0	168983	0.0492	0.0505	0.0131
0.0023	30.0	174810	0.0536	0.0487	0.0120
0.0022	31.0	180637	0.0724	0.0539	0.0120
0.0022	32.0	186464	0.0555	0.0506	0.0121
0.0019	33.0	192291	0.0718	0.0477	0.0114
0.0019	34.0	198118	0.0662	0.0520	0.0124
0.0018	35.0	203945	0.0712	0.0487	0.0109
0.0016	36.0	209772	0.0578	0.0489	0.0116
0.0014	37.0	215599	0.0633	0.0537	0.0117

Framework versions

• Transformers 4.47.0
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.21.0