w2v-bert-2.0-hausa_naijavoices_1m

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the CLEAR-GLOBAL/NAIJAVOICES_1M - NA dataset. It achieves the following results on the evaluation set:

• Loss: 1.5621
• Wer: 1.0108
• Cer: 0.4556

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: 3e-05
• train_batch_size: 160
• eval_batch_size: 160
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 320
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1500000.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.8167	100.0	100	5.0671	1.0142	1.3987
2.7438	200.0	200	4.9565	1.0269	1.3063
2.6019	300.0	300	4.7672	1.0480	1.1211
2.3831	400.0	400	4.4957	1.0348	0.8779
2.1047	500.0	500	4.0627	0.9868	0.8782
1.8364	600.0	600	3.6321	0.9962	0.9904
1.6452	700.0	700	3.2890	0.9978	0.9951
1.5101	800.0	800	3.0365	0.9900	0.9676
1.4058	900.0	900	2.8747	0.9898	0.9242
1.3447	1000.0	1000	2.7941	0.9895	0.8413
1.2893	1100.0	1100	2.7511	0.9894	0.7897
1.225	1200.0	1200	2.6778	0.9911	0.7638
1.1437	1300.0	1300	2.5807	1.0164	0.7204
1.0582	1400.0	1400	2.4711	1.0629	0.6844
0.9647	1500.0	1500	2.3571	1.1223	0.6542
0.8616	1600.0	1600	2.2293	1.1496	0.6331
0.7263	1700.0	1700	2.1078	1.1669	0.6155
0.6343	1800.0	1800	1.9905	1.1426	0.5992
0.5373	1900.0	1900	1.9008	1.1520	0.5821
0.4254	2000.0	2000	1.8263	1.1235	0.5649
0.3388	2100.0	2100	1.7375	1.0959	0.5414
0.263	2200.0	2200	1.6725	1.0716	0.5187
0.1962	2300.0	2300	1.6141	1.0550	0.4970
0.1423	2400.0	2400	1.5796	1.0406	0.4798
0.1017	2500.0	2500	1.5637	1.0233	0.4656
0.0739	2600.0	2600	1.5620	1.0111	0.4557
0.055	2700.0	2700	1.5684	1.0014	0.4463
0.0356	2800.0	2800	1.5855	0.9978	0.4380
0.0275	2900.0	2900	1.6100	0.9833	0.4321
0.0217	3000.0	3000	1.6374	0.9823	0.4291
0.0178	3100.0	3100	1.6611	0.9802	0.4242

Framework versions

• Transformers 4.48.1
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.1