w2v-bert-2.0-igbo_naijavoices_100h

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the NAIJAVOICES_IGBO_100H - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.3166
• Wer: 0.3077
• Cer: 0.1314

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
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 320
• total_eval_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
0.638	3.4483	1000	0.6548	0.5546	0.2086
0.4035	6.8966	2000	0.4278	0.3949	0.1574
0.5526	10.3448	3000	0.3694	0.3530	0.1461
0.305	13.7931	4000	0.3367	0.3357	0.1392
0.2602	17.2414	5000	0.3230	0.3252	0.1366
0.4161	20.6897	6000	0.3192	0.3226	0.1360
0.2367	24.1379	7000	0.3196	0.3192	0.1345
0.2463	27.5862	8000	0.3153	0.3089	0.1324
0.1819	31.0345	9000	0.3187	0.3034	0.1334
0.1852	34.4828	10000	0.3292	0.3067	0.1326
0.1581	37.9310	11000	0.3510	0.2997	0.1310
0.1455	41.3793	12000	0.3522	0.3080	0.1348
0.0998	44.8276	13000	0.3933	0.3094	0.1373

Framework versions

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