whisper-small-CV_Fleurs_AMMI_ALFFA-sw-10hrs-v1

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: 0.9490
• Wer: 0.2851
• Cer: 0.1098

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: 5e-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: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.7806	0.9986	360	0.8995	0.6431	0.2391
1.3216	2.0	721	0.6455	0.4377	0.1689
0.7711	2.9986	1081	0.5796	0.4257	0.1831
0.439	4.0	1442	0.5872	0.4005	0.1697
0.275	4.9986	1802	0.5826	0.3129	0.1115
0.2006	6.0	2163	0.6350	0.3191	0.1166
0.1783	6.9986	2523	0.6342	0.3165	0.1108
0.1629	8.0	2884	0.6669	0.3099	0.1127
0.1564	8.9986	3244	0.6764	0.3390	0.1387
0.1575	10.0	3605	0.7052	0.3367	0.1351
0.1524	10.9986	3965	0.7088	0.3116	0.1149
0.128	12.0	4326	0.7365	0.3661	0.1603
0.1148	12.9986	4686	0.7283	0.3089	0.1132
0.0946	14.0	5047	0.7364	0.3046	0.1118
0.0826	14.9986	5407	0.7805	0.2963	0.1105
0.0751	16.0	5768	0.7608	0.3273	0.1291
0.0697	16.9986	6128	0.7748	0.2941	0.1089
0.0659	18.0	6489	0.7887	0.3064	0.1142
0.0547	18.9986	6849	0.7944	0.2943	0.1124
0.0531	20.0	7210	0.8065	0.2921	0.1094
0.0495	20.9986	7570	0.7981	0.2897	0.1056
0.0481	22.0	7931	0.8201	0.2927	0.1089
0.0437	22.9986	8291	0.8175	0.2958	0.1126
0.0392	24.0	8652	0.8470	0.3106	0.1214
0.0379	24.9986	9012	0.8203	0.2965	0.1143
0.0332	26.0	9373	0.8315	0.2848	0.1062
0.0335	26.9986	9733	0.8485	0.2857	0.1040
0.0296	28.0	10094	0.8591	0.2997	0.1154
0.0267	28.9986	10454	0.8672	0.2910	0.1084
0.0291	30.0	10815	0.8552	0.2905	0.1069
0.0275	30.9986	11175	0.8441	0.2901	0.1081
0.0248	32.0	11536	0.8693	0.2897	0.1107
0.0233	32.9986	11896	0.8582	0.2830	0.1037
0.0208	34.0	12257	0.8740	0.2911	0.1112
0.0186	34.9986	12617	0.8843	0.2832	0.1061
0.0187	36.0	12978	0.8962	0.2877	0.1056
0.0186	36.9986	13338	0.9004	0.2855	0.1069
0.0174	38.0	13699	0.8975	0.2829	0.1059
0.0151	38.9986	14059	0.8970	0.2879	0.1081
0.0146	40.0	14420	0.9029	0.2877	0.1075
0.014	40.9986	14780	0.9081	0.2815	0.1038
0.0155	42.0	15141	0.8880	0.2774	0.1057
0.0138	42.9986	15501	0.9054	0.2810	0.1048
0.0136	44.0	15862	0.9054	0.2807	0.1078
0.013	44.9986	16222	0.9033	0.2810	0.1066
0.011	46.0	16583	0.9044	0.2833	0.1097
0.0105	46.9986	16943	0.9064	0.2796	0.1036
0.0101	48.0	17304	0.9049	0.2823	0.1058
0.0072	48.9986	17664	0.9250	0.2870	0.1086
0.0116	50.0	18025	0.9556	0.2729	0.1001
0.0094	50.9986	18385	0.9187	0.2820	0.1073
0.0069	52.0	18746	0.9249	0.2828	0.1092
0.0077	52.9986	19106	0.9443	0.2891	0.1139
0.0088	54.0	19467	0.9336	0.2888	0.1127
0.0076	54.9986	19827	0.9416	0.2831	0.1074
0.0064	56.0	20188	0.9477	0.2790	0.1058
0.0055	56.9986	20548	0.9273	0.2772	0.1050
0.005	58.0	20909	0.9315	0.2781	0.1056
0.0055	58.9986	21269	0.9505	0.2855	0.1094
0.0064	60.0	21630	0.9490	0.2851	0.1098

Framework versions

• Transformers 4.46.1
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.1