luganda_wav2vec2_ctc_tokenizer

This model is a fine-tuned version of facebook/wav2vec2-base on the common_voice_7_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5588
• Wer: 0.5609

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: 0.0001
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.1365	2.4	500	1.9598	1.0
0.5695	4.81	1000	0.5853	0.7329
0.176	7.21	1500	0.5381	0.6747
0.0845	9.62	2000	0.5128	0.6270
0.0424	12.02	2500	0.4651	0.6014
0.0127	14.42	3000	0.5395	0.6049
-0.0063	16.83	3500	0.5169	0.5842
-0.0212	19.23	4000	0.4990	0.5833
-0.0336	21.63	4500	0.5318	0.5680
-0.0424	24.04	5000	0.5465	0.5702
-0.0495	26.44	5500	0.5541	0.5637
-0.0565	28.85	6000	0.5588	0.5609

Framework versions

• Transformers 4.37.2
• Pytorch 2.1.0+cu121
• Datasets 2.17.1
• Tokenizers 0.15.2