wavlm-base-ug-combined

This model is a fine-tuned version of microsoft/wavlm-base on the AJIKADEV/UGANDAN-ENGLISH - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.1281
• Wer: 0.1548

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: 0.0003
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500.0
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5871	0.8042	1000	0.4082	0.4143
0.3352	1.6080	2000	0.2659	0.2979
0.2262	2.4117	3000	0.2189	0.2459
0.1922	3.2155	4000	0.1928	0.2205
0.1527	4.0193	5000	0.1667	0.2024
0.1152	4.8235	6000	0.1525	0.1893
0.0932	5.6273	7000	0.1435	0.1775
0.0721	6.4310	8000	0.1365	0.1672
0.0559	7.2348	9000	0.1335	0.1582
0.0483	8.0386	10000	0.1281	0.1545

Framework versions

• Transformers 5.0.0.dev0
• Pytorch 2.9.1+cu128
• Datasets 3.6.0
• Tokenizers 0.22.1