w2vbert-lingala-waxal-punct-v2

This model is a fine-tuned version of sulaimank/w2vbert-lingala-sd3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0860
• Wer Keep: 0.1140
• Cer Keep: 0.0326
• Zindi Keep: 0.9267
• Wer Strip: 0.0619
• Zindi Strip: 0.9607
• Zindi Lower: 0.9688

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: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.98) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 0.1
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Keep	Cer Keep	Zindi Keep	Wer Strip	Zindi Strip	Zindi Lower
0.5347	0.0812	200	0.0756	0.1039	0.0298	0.9332	0.0530	0.9664	0.9745
0.7679	0.1624	400	0.0770	0.1054	0.0295	0.9325	0.0550	0.9651	0.9734
0.3571	0.2436	600	0.0799	0.1073	0.0290	0.9318	0.0567	0.9641	0.9725
0.4512	0.3248	800	0.0816	0.1104	0.0320	0.9288	0.0602	0.9618	0.9703
0.6859	0.4060	1000	0.0856	0.1102	0.0316	0.9291	0.0601	0.9618	0.9700
0.7610	0.4872	1200	0.0889	0.1138	0.0316	0.9273	0.0614	0.9610	0.9692
0.6022	0.5684	1400	0.0800	0.1107	0.0310	0.9291	0.0600	0.9620	0.9701
0.3510	0.6496	1600	0.0851	0.1140	0.0323	0.9268	0.0631	0.9600	0.9683
0.4350	0.7308	1800	0.0860	0.1140	0.0326	0.9267	0.0619	0.9607	0.9688

Framework versions

• Transformers 5.14.1
• Pytorch 2.13.0+cu130
• Datasets 3.6.0
• Tokenizers 0.22.2