wav2vec-xlsr-cv-grain-lg_grn_only_v2

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0604
• Wer: 0.0276
• Cer: 0.0085

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: 0.0003
• train_batch_size: 24
• eval_batch_size: 12
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 48
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
6.8998	0.9984	321	2.7793	1.0	0.8727
3.2905	2.0	643	0.8365	0.9015	0.2478
1.26	2.9984	964	0.3066	0.4268	0.0856
0.6344	4.0	1286	0.1856	0.2137	0.0451
0.4164	4.9984	1607	0.1513	0.1649	0.0364
0.3006	6.0	1929	0.1271	0.1274	0.0285
0.2414	6.9984	2250	0.1111	0.1083	0.0251
0.2035	8.0	2572	0.1076	0.0992	0.0228
0.169	8.9984	2893	0.1076	0.0931	0.0213
0.1501	10.0	3215	0.1007	0.0920	0.0213
0.1291	10.9984	3536	0.0892	0.0772	0.0185
0.1122	12.0	3858	0.0917	0.0746	0.0180
0.1053	12.9984	4179	0.0903	0.0707	0.0173
0.0972	14.0	4501	0.0863	0.0673	0.0164
0.0847	14.9984	4822	0.0849	0.0616	0.0157
0.0754	16.0	5144	0.0870	0.0657	0.0158
0.0751	16.9984	5465	0.0830	0.0610	0.0154
0.0722	18.0	5787	0.0922	0.0621	0.0159
0.0665	18.9984	6108	0.0784	0.0601	0.0153
0.0634	20.0	6430	0.0856	0.0545	0.0146
0.0601	20.9984	6751	0.0881	0.0584	0.0151
0.0545	22.0	7073	0.0876	0.0558	0.0144
0.0503	22.9984	7394	0.0815	0.0523	0.0137
0.0511	24.0	7716	0.0842	0.0521	0.0140
0.0477	24.9984	8037	0.0808	0.0532	0.0151
0.0433	26.0	8359	0.0770	0.0482	0.0125
0.0441	26.9984	8680	0.0803	0.0510	0.0137
0.0424	28.0	9002	0.0771	0.0460	0.0123
0.0373	28.9984	9323	0.0727	0.0462	0.0122
0.0376	30.0	9645	0.0768	0.0525	0.0134
0.0325	30.9984	9966	0.0801	0.0508	0.0134
0.0371	32.0	10288	0.0714	0.0445	0.0118
0.0339	32.9984	10609	0.0738	0.0458	0.0122
0.0329	34.0	10931	0.0672	0.0388	0.0104
0.0294	34.9984	11252	0.0750	0.0408	0.0113
0.0322	36.0	11574	0.0768	0.0423	0.0117
0.028	36.9984	11895	0.0735	0.0386	0.0117
0.0279	38.0	12217	0.0756	0.0414	0.0122
0.0259	38.9984	12538	0.0842	0.0495	0.0135
0.0273	40.0	12860	0.0775	0.0456	0.0131
0.026	40.9984	13181	0.0729	0.0427	0.0119
0.0247	42.0	13503	0.0728	0.0410	0.0115
0.0247	42.9984	13824	0.0709	0.0430	0.0118
0.023	44.0	14146	0.0632	0.0362	0.0101
0.0206	44.9984	14467	0.0675	0.0347	0.0106
0.0203	46.0	14789	0.0750	0.0419	0.0125
0.0215	46.9984	15110	0.0644	0.0358	0.0104
0.0172	48.0	15432	0.0693	0.0332	0.0098
0.0191	48.9984	15753	0.0694	0.0341	0.0102
0.0175	50.0	16075	0.0716	0.0369	0.0108
0.018	50.9984	16396	0.0635	0.0351	0.0101
0.0162	52.0	16718	0.0711	0.0382	0.0106
0.0167	52.9984	17039	0.0605	0.0343	0.0097
0.0173	54.0	17361	0.0699	0.0321	0.0097
0.0157	54.9984	17682	0.0726	0.0330	0.0100
0.0128	56.0	18004	0.0693	0.0323	0.0096
0.0169	56.9984	18325	0.0602	0.0306	0.0092
0.014	58.0	18647	0.0638	0.0332	0.0097
0.0133	58.9984	18968	0.0630	0.0325	0.0097
0.0151	60.0	19290	0.0645	0.0328	0.0098
0.0137	60.9984	19611	0.0642	0.0351	0.0098
0.0135	62.0	19933	0.0569	0.0284	0.0084
0.0119	62.9984	20254	0.0595	0.0308	0.0088
0.011	64.0	20576	0.0601	0.0263	0.0086
0.0113	64.9984	20897	0.0639	0.0282	0.0090
0.0125	66.0	21219	0.0588	0.0291	0.0090
0.0103	66.9984	21540	0.0632	0.0289	0.0090
0.0094	68.0	21862	0.0600	0.0282	0.0087
0.0098	68.9984	22183	0.0615	0.0278	0.0085
0.0089	70.0	22505	0.0598	0.0278	0.0084
0.0105	70.9984	22826	0.0611	0.0291	0.0081
0.0083	72.0	23148	0.0623	0.0293	0.0084
0.0092	72.9984	23469	0.0590	0.0302	0.0090
0.0068	74.0	23791	0.0604	0.0276	0.0085

Framework versions

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