whisper-small-lg-CV_GRAIN-v3

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: 0.0556
Wer: 0.0383
Cer: 0.0095

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
2.158	1.0	1385	0.5563	1.1119	0.4090
0.5591	2.0	2770	0.2041	0.9345	0.3983
0.2976	3.0	4155	0.1080	0.8668	0.3299
0.174	4.0	5540	0.0709	0.9049	0.4095
0.0992	5.0	6925	0.0586	0.8077	0.3630
0.0552	6.0	8310	0.0554	0.5285	0.2370
0.0327	7.0	9695	0.0555	0.4657	0.2091
0.0225	8.0	11080	0.0616	0.2815	0.1215
0.0167	9.0	12465	0.0522	0.1662	0.0634
0.0119	10.0	13850	0.0577	0.1821	0.0806
0.0087	11.0	15235	0.0544	0.1336	0.0596
0.0064	12.0	16620	0.0552	0.1255	0.0492
0.0055	13.0	18005	0.0597	0.1871	0.0867
0.0044	14.0	19390	0.0567	0.0992	0.0360
0.0043	15.0	20775	0.0618	0.0963	0.0321
0.0031	16.0	22160	0.0587	0.0868	0.0245
0.003	17.0	23545	0.0572	0.0930	0.0288
0.0023	18.0	24930	0.0629	0.0918	0.0288
0.0026	19.0	26315	0.0565	0.0860	0.0289
0.0025	20.0	27700	0.0550	0.0754	0.0247
0.002	21.0	29085	0.0595	0.0675	0.0216
0.0023	22.0	30470	0.0602	0.0700	0.0214
0.0018	23.0	31855	0.0594	0.0646	0.0176
0.002	24.0	33240	0.0594	0.0684	0.0195
0.0016	25.0	34625	0.0600	0.0704	0.0246
0.0016	26.0	36010	0.0518	0.0619	0.0177
0.0012	27.0	37395	0.0538	0.0555	0.0163
0.0012	28.0	38780	0.0522	0.0595	0.0158
0.0013	29.0	40165	0.0584	0.0566	0.0152
0.0011	30.0	41550	0.0508	0.0512	0.0137
0.0012	31.0	42935	0.0649	0.0566	0.0136
0.0009	32.0	44320	0.0615	0.0615	0.0170
0.0011	33.0	45705	0.0562	0.0522	0.0140
0.0009	34.0	47090	0.0596	0.0539	0.0158
0.001	35.0	48475	0.0577	0.0539	0.0143
0.0006	36.0	49860	0.0441	0.0493	0.0140
0.0007	37.0	51245	0.0739	0.0561	0.0148
0.0008	38.0	52630	0.0490	0.0501	0.0144
0.0006	39.0	54015	0.0530	0.0503	0.0133
0.0005	40.0	55400	0.0650	0.0493	0.0129
0.0006	41.0	56785	0.0624	0.0514	0.0135
0.0006	42.0	58170	0.0629	0.0518	0.0135
0.0006	43.0	59555	0.0703	0.0524	0.0139
0.0008	44.0	60940	0.0552	0.0481	0.0147
0.0004	45.0	62325	0.0588	0.0462	0.0118
0.0003	46.0	63710	0.0540	0.0423	0.0106
0.0003	47.0	65095	0.0550	0.0483	0.0125
0.0002	48.0	66480	0.0552	0.0472	0.0129
0.0003	49.0	67865	0.0615	0.0435	0.0107
0.0004	50.0	69250	0.0522	0.0456	0.0125
0.0001	51.0	70635	0.0535	0.0416	0.0116
0.0002	52.0	72020	0.0612	0.0433	0.0105
0.0002	53.0	73405	0.0530	0.0421	0.0124
0.0002	54.0	74790	0.0575	0.0435	0.0105
0.0001	55.0	76175	0.0466	0.0402	0.0117
0.0003	56.0	77560	0.0499	0.0447	0.0123
0.0002	57.0	78945	0.0510	0.0435	0.0116
0.0001	58.0	80330	0.0509	0.0447	0.0124
0.0001	59.0	81715	0.0472	0.0400	0.0105
0.0	60.0	83100	0.0554	0.0416	0.0113
0.0001	61.0	84485	0.0516	0.0524	0.0186
0.0002	62.0	85870	0.0541	0.0437	0.0112
0.0001	63.0	87255	0.0533	0.0410	0.0107
0.0001	64.0	88640	0.0480	0.0406	0.0112
0.0	65.0	90025	0.0580	0.0435	0.0115
0.0001	66.0	91410	0.0492	0.0447	0.0117
0.0001	67.0	92795	0.0554	0.0381	0.0095
0.0	68.0	94180	0.0545	0.0385	0.0099
0.0	69.0	95565	0.0510	0.0363	0.0096
0.0	70.0	96950	0.0534	0.0389	0.0101
0.0	71.0	98335	0.0537	0.0394	0.0104
0.0	72.0	99720	0.0549	0.0392	0.0103
0.0	73.0	101105	0.0541	0.0383	0.0098
0.0	74.0	102490	0.0544	0.0389	0.0099
0.0	75.0	103875	0.0543	0.0392	0.0099
0.0	76.0	105260	0.0548	0.0390	0.0097
0.0	77.0	106645	0.0552	0.0389	0.0096
0.0	78.0	108030	0.0554	0.0385	0.0095
0.0	79.0	109415	0.0556	0.0383	0.0095
0.0	80.0	110800	0.0556	0.0383	0.0095

Framework versions

Transformers 4.47.0
Pytorch 2.1.0+cu118
Datasets 3.1.0
Tokenizers 0.21.0

sulaimank/whisper-small-lg-GRAIN-v3

whisper-small-lg-CV_GRAIN-v3

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Seed this model