w2vbert-luganda-waxal-punct-v2

This model is a fine-tuned version of sulaimank/w2vbert-luganda-waxal-stage2 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1568
• Wer Keep: 0.1779
• Cer Keep: 0.0389
• Zindi Keep: 0.8916
• Wer Strip: 0.0962
• Cer Strip: 0.0218
• Zindi Strip: 0.9410

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Keep	Cer Keep	Zindi Keep	Wer Strip	Cer Strip	Zindi Strip
0.3557	0.2516	400	0.1668	0.2007	0.0450	0.8772	0.1125	0.0259	0.9308
0.2105	0.5032	800	0.1477	0.1840	0.0408	0.8876	0.1022	0.0233	0.9372
0.2031	0.7548	1200	0.1340	0.1808	0.0404	0.8894	0.1002	0.0228	0.9385
0.1561	1.0063	1600	0.1328	0.1783	0.0398	0.8909	0.0974	0.0220	0.9403
0.1456	1.2579	2000	0.1397	0.1767	0.0384	0.8925	0.0959	0.0219	0.9411
0.1168	1.5095	2400	0.1415	0.1761	0.0382	0.8928	0.0948	0.0221	0.9416
0.1438	1.7611	2800	0.1419	0.1773	0.0373	0.8927	0.0967	0.0220	0.9406
0.0890	2.0126	3200	0.1554	0.1759	0.0374	0.8934	0.0952	0.0217	0.9416
0.0892	2.2642	3600	0.1571	0.1782	0.0389	0.8914	0.0962	0.0220	0.9409
0.0998	2.5158	4000	0.1529	0.1776	0.0386	0.8919	0.0965	0.0217	0.9409
0.0761	2.7674	4400	0.1574	0.1779	0.0388	0.8916	0.0974	0.0220	0.9403
0.0799	3.0	4770	0.1568	0.1779	0.0389	0.8916	0.0962	0.0218	0.9410

Framework versions

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