wav2vec-300m-max18-WF-epoch-16-batch-8

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the IndabaxSenegal/asr-wolof-dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.9861
• Wer: 0.3930

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 16
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.2704	0.37	500	7.9492	1.0
3.19	0.75	1000	5.3047	1.0
2.7353	1.12	1500	2.1307	0.8724
2.058	1.49	2000	1.5998	0.6999
1.7582	1.87	2500	1.3994	0.6739
1.6432	2.24	3000	1.2822	0.5821
1.4893	2.61	3500	1.2704	0.5579
1.5018	2.99	4000	1.2324	0.5374
1.3868	3.36	4500	1.1459	0.5254
1.3156	3.73	5000	1.1376	0.5073
1.3179	4.11	5500	1.2034	0.5187
1.2562	4.48	6000	1.1535	0.4853
1.199	4.85	6500	1.0877	0.4797
1.1622	5.23	7000	1.1037	0.4693
1.1349	5.6	7500	1.1353	0.4625
1.1241	5.97	8000	1.0888	0.4491
1.0526	6.35	8500	1.0463	0.4508
1.0653	6.72	9000	1.2282	0.4432
1.0466	7.09	9500	1.1014	0.4330
1.009	7.47	10000	1.0511	0.4281
0.9773	7.84	10500	1.0790	0.4221
0.9662	8.22	11000	1.0327	0.4252
0.9973	8.59	11500	1.0801	0.4230
0.928	8.96	12000	1.0048	0.4179
0.8971	9.34	12500	1.0386	0.4153
0.919	9.71	13000	1.0200	0.4190
0.9131	10.08	13500	1.0283	0.4065
0.895	10.46	14000	1.0359	0.4042
0.8723	10.83	14500	0.9785	0.4096
0.8622	11.2	15000	1.0242	0.4044
0.8346	11.58	15500	1.0548	0.4083
0.8547	11.95	16000	0.9884	0.3990
0.8355	12.32	16500	1.0303	0.3959
0.846	12.7	17000	1.0079	0.3973
0.812	13.07	17500	1.0344	0.4027
0.8028	13.44	18000	0.9861	0.3930
0.8028	13.82	18500	1.0077	0.3954
0.7908	14.19	19000	0.9927	0.3949
0.7992	14.56	19500	0.9880	0.3957
0.7995	14.94	20000	0.9715	0.3931
0.8056	15.31	20500	0.9837	0.3966
0.7826	15.68	21000	0.9737	0.3973

Framework versions

• Transformers 4.35.2
• Pytorch 2.4.0
• Datasets 2.15.0
• Tokenizers 0.15.2