w2v-bert-2.0-yoruba_naijavoices_100h

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the NAIJAVOICES_YORUBA_100H - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.5647
• Wer: 0.5228
• Cer: 0.2858

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: 250.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.8442	3.2051	1000	0.8847	0.6790	0.3471
0.6917	6.4103	2000	0.7064	0.6078	0.3204
0.7078	9.6154	3000	0.6668	0.5840	0.3096
0.6132	12.8205	4000	0.6308	0.5640	0.3052
0.577	16.0256	5000	0.6232	0.5551	0.3010
0.5954	19.2308	6000	0.6014	0.5390	0.2942
0.5286	22.4359	7000	0.6020	0.5449	0.2986
0.7866	25.6410	8000	0.5781	0.5326	0.2904
0.4878	28.8462	9000	0.5705	0.5100	0.2935
0.4091	32.0513	10000	0.5647	0.5218	0.2857
0.3913	35.2564	11000	0.5915	0.5248	0.2921
0.3878	38.4615	12000	0.6052	0.5296	0.2998
0.364	41.6667	13000	0.6151	0.4967	0.2802
0.3021	44.8718	14000	0.6075	0.4997	0.2861
0.2587	48.0769	15000	0.6518	0.5329	0.3016

Framework versions

• Transformers 4.48.1
• Pytorch 2.7.1+cu126
• Datasets 4.0.0
• Tokenizers 0.21.2