wav2vec2-xls-r-300m-CV-Fleurs-lg-5hrs-v7

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: 0.7601
• Wer: 0.4290
• Cer: 0.0891

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: 0.0001
• train_batch_size: 4
• eval_batch_size: 2
• seed: 42
• 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: 70
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.767	1.0	515	2.9466	1.0	1.0
2.5111	2.0	1030	1.1124	0.9072	0.2504
1.0272	3.0	1545	0.6239	0.7472	0.1651
0.7225	4.0	2060	0.5120	0.6347	0.1391
0.5936	5.0	2575	0.4842	0.6171	0.1380
0.5104	6.0	3090	0.4519	0.6032	0.1387
0.4486	7.0	3605	0.4516	0.5645	0.1177
0.3944	8.0	4120	0.4659	0.5373	0.1146
0.3587	9.0	4635	0.4384	0.5231	0.1092
0.3259	10.0	5150	0.4401	0.5228	0.1155
0.2929	11.0	5665	0.4667	0.5164	0.1078
0.278	12.0	6180	0.4671	0.5099	0.1074
0.2597	13.0	6695	0.4737	0.5001	0.1059
0.2428	14.0	7210	0.4900	0.4987	0.1043
0.2277	15.0	7725	0.5479	0.5282	0.1077
0.2126	16.0	8240	0.5342	0.5019	0.1052
0.1955	17.0	8755	0.5155	0.4901	0.1014
0.1892	18.0	9270	0.5369	0.4889	0.1011
0.1733	19.0	9785	0.5470	0.4821	0.1016
0.1736	20.0	10300	0.5909	0.4833	0.1003
0.1594	21.0	10815	0.5675	0.4851	0.1017
0.1576	22.0	11330	0.6015	0.4846	0.1016
0.1511	23.0	11845	0.5394	0.4690	0.0983
0.1419	24.0	12360	0.6172	0.4848	0.0996
0.1359	25.0	12875	0.6323	0.4782	0.1005
0.1411	26.0	13390	0.5898	0.4739	0.0992
0.1279	27.0	13905	0.6021	0.4792	0.0994
0.1152	28.0	14420	0.6077	0.4713	0.0981
0.1187	29.0	14935	0.6468	0.4702	0.0987
0.109	30.0	15450	0.6424	0.4731	0.0981
0.1054	31.0	15965	0.6620	0.4729	0.0985
0.1105	32.0	16480	0.6593	0.4607	0.0972
0.1011	33.0	16995	0.6920	0.4724	0.0974
0.0972	34.0	17510	0.6946	0.4719	0.0975
0.1009	35.0	18025	0.6664	0.4652	0.0962
0.0925	36.0	18540	0.6852	0.4664	0.0966
0.0907	37.0	19055	0.6871	0.4619	0.0971
0.0931	38.0	19570	0.6975	0.4512	0.0944
0.0859	39.0	20085	0.6803	0.4528	0.0950
0.0844	40.0	20600	0.7284	0.4646	0.0966
0.0832	41.0	21115	0.7195	0.4502	0.0940
0.0752	42.0	21630	0.6899	0.4517	0.0944
0.0787	43.0	22145	0.7047	0.4494	0.0939
0.0762	44.0	22660	0.7115	0.4531	0.0940
0.0709	45.0	23175	0.7336	0.4522	0.0932
0.0697	46.0	23690	0.7023	0.4468	0.0921
0.0737	47.0	24205	0.7567	0.4470	0.0929
0.0684	48.0	24720	0.7335	0.4437	0.0912
0.0677	49.0	25235	0.7230	0.4442	0.0916
0.0704	50.0	25750	0.7172	0.4510	0.0928
0.0637	51.0	26265	0.7513	0.4450	0.0927
0.0632	52.0	26780	0.7554	0.4548	0.0938
0.0562	53.0	27295	0.7654	0.4508	0.0938
0.0576	54.0	27810	0.7444	0.4389	0.0914
0.055	55.0	28325	0.7285	0.4426	0.0915
0.0555	56.0	28840	0.7418	0.4405	0.0915
0.0633	57.0	29355	0.7509	0.4412	0.0918
0.0533	58.0	29870	0.7486	0.4392	0.0916
0.0554	59.0	30385	0.7415	0.4420	0.0918
0.0555	60.0	30900	0.7485	0.4413	0.0916
0.0524	61.0	31415	0.7463	0.4372	0.0906
0.0476	62.0	31930	0.7501	0.4337	0.0901
0.0503	63.0	32445	0.7474	0.4309	0.0898
0.0458	64.0	32960	0.7603	0.4302	0.0897
0.0495	65.0	33475	0.7558	0.4309	0.0894
0.0518	66.0	33990	0.7585	0.4282	0.0887
0.051	67.0	34505	0.7586	0.4297	0.0893
0.0483	68.0	35020	0.7586	0.4291	0.0893
0.0445	69.0	35535	0.7604	0.4297	0.0893
0.0437	70.0	36050	0.7601	0.4290	0.0891

Framework versions

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