wav2vec2_xls_r_300m_DigitalUmuganda_Afrivoice_Shona_5hr_v2

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: 0.8167
• Wer: 0.5757
• Cer: 0.1266

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
9.1759	1.8182	200	3.6548	1.0	1.0
3.176	3.6364	400	2.9716	1.0	1.0
2.8224	5.4545	600	2.4223	1.0	0.7811
0.8897	7.2727	800	1.1037	0.9577	0.3723
0.3633	9.0909	1000	0.5800	0.7405	0.1671
0.2495	10.9091	1200	0.6582	0.8157	0.2182
0.1947	12.7273	1400	0.6504	0.7687	0.2051
0.1581	14.5455	1600	0.3927	0.5197	0.0987
0.1201	16.3636	1800	0.4514	0.5761	0.1127
0.1055	18.1818	2000	0.5996	0.7252	0.1701
0.0908	20.0	2200	0.6709	0.7622	0.1981
0.0769	21.8182	2400	0.5096	0.6284	0.1309
0.0654	23.6364	2600	0.5252	0.6187	0.1273
0.0571	25.4545	2800	0.5280	0.6206	0.1275
0.0531	27.2727	3000	0.5062	0.5873	0.1223
0.0465	29.0909	3200	0.5311	0.5919	0.1249
0.0448	30.9091	3400	0.5093	0.5725	0.1165
0.042	32.7273	3600	0.5431	0.5953	0.1248
0.0398	34.5455	3800	0.5109	0.5659	0.1123
0.0347	36.3636	4000	0.5269	0.5885	0.1194
0.0323	38.1818	4200	0.5526	0.5934	0.1205
0.0341	40.0	4400	0.4783	0.5015	0.0960
0.0312	41.8182	4600	0.5495	0.5696	0.1142
0.0298	43.6364	4800	0.5522	0.5725	0.1154
0.0274	45.4545	5000	0.5945	0.5807	0.1212
0.0261	47.2727	5200	0.5271	0.5501	0.1081
0.0267	49.0909	5400	0.5689	0.5722	0.1145
0.026	50.9091	5600	0.5364	0.5350	0.1021
0.0238	52.7273	5800	0.5312	0.5151	0.0969
0.0236	54.5455	6000	0.5327	0.5338	0.1028
0.0223	56.3636	6200	0.5174	0.5248	0.0995
0.0217	58.1818	6400	0.5339	0.5255	0.1004
0.0218	60.0	6600	0.5461	0.5375	0.1036
0.02	61.8182	6800	0.5194	0.4954	0.0909
0.0194	63.6364	7000	0.5458	0.5202	0.0983
0.0188	65.4545	7200	0.5133	0.4869	0.0900
0.0189	67.2727	7400	0.5141	0.4910	0.0901
0.0179	69.0909	7600	0.5300	0.4864	0.0881
0.017	70.9091	7800	0.5256	0.4759	0.0861
0.017	72.7273	8000	0.5221	0.4713	0.0859
0.0153	74.5455	8200	0.5419	0.4803	0.0879
0.0172	76.3636	8400	0.5395	0.5073	0.0940
0.0156	78.1818	8600	0.5519	0.5039	0.0941
0.0153	80.0	8800	0.5436	0.4971	0.0904
0.0155	81.8182	9000	0.5477	0.4983	0.0919
0.014	83.6364	9200	0.5352	0.4905	0.0887
0.0144	85.4545	9400	0.5328	0.4866	0.0873
0.0136	87.2727	9600	0.5403	0.4900	0.0888
0.0136	89.0909	9800	0.5457	0.4939	0.0899
0.0133	90.9091	10000	0.5445	0.4927	0.0898
0.0142	92.7273	10200	0.5413	0.4934	0.0902
0.0149	94.5455	10400	0.5441	0.4959	0.0910
0.0138	96.3636	10600	0.5434	0.4951	0.0905
0.0136	98.1818	10800	0.5438	0.4961	0.0906
0.0141	100.0	11000	0.5439	0.4951	0.0906

Framework versions

• Transformers 4.44.0
• Pytorch 2.2.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1