wav2vec2-xls-r-Wolof-10-hours-kallaama-dataset

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

• Loss: 1.4430
• Wer: 0.4016
• Cer: 0.1944

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
4.7569	2.2599	400	3.0479	1.0	1.0
2.8776	4.5198	800	1.9229	0.9892	0.6359
1.9358	6.7797	1200	1.4536	0.7002	0.3496
1.5228	9.0395	1600	1.1791	0.5682	0.2775
1.2522	11.2994	2000	1.0833	0.5115	0.2410
1.1025	13.5593	2400	1.0679	0.4835	0.2284
0.9742	15.8192	2800	1.0346	0.4628	0.2212
0.8979	18.0791	3200	1.0595	0.4509	0.2147
0.8198	20.3390	3600	1.0954	0.4344	0.2061
0.7734	22.5989	4000	1.0795	0.4334	0.2091
0.6949	24.8588	4400	1.1170	0.4253	0.2037
0.6541	27.1186	4800	1.1553	0.4327	0.2067
0.6012	29.3785	5200	1.1821	0.4188	0.2015
0.5542	31.6384	5600	1.2288	0.4226	0.2032
0.5251	33.8983	6000	1.2887	0.4199	0.2017
0.4779	36.1582	6400	1.3335	0.4141	0.2001
0.4524	38.4181	6800	1.3469	0.4132	0.1996
0.4257	40.6780	7200	1.3576	0.4101	0.1978
0.3887	42.9379	7600	1.3911	0.4074	0.1958
0.3748	45.1977	8000	1.4059	0.4062	0.1958
0.354	47.4576	8400	1.4341	0.4021	0.1944
0.3485	49.7175	8800	1.4430	0.4016	0.1944

Framework versions

• Transformers 4.44.1
• Pytorch 2.1.0+cu118
• Datasets 2.17.0
• Tokenizers 0.19.1