wav2vec2-large-xls-r-may23-luganda-colab

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.7210
• Wer: 0.5021

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0539	7.77	400	0.6641	0.5738
0.0725	15.53	800	0.6735	0.5932
0.058	23.3	1200	0.6754	0.5751
0.0517	31.07	1600	0.6591	0.5901
0.0437	38.83	2000	0.7140	0.5658
0.0366	46.6	2400	0.7154	0.5602
0.0295	54.37	2800	0.6942	0.5140
0.0251	62.14	3200	0.7095	0.5204
0.0191	69.9	3600	0.7459	0.5267
0.0157	77.67	4000	0.6825	0.5155
0.0126	85.44	4400	0.7197	0.5135
0.0098	93.2	4800	0.7210	0.5021

Framework versions

• Transformers 4.29.2
• Pytorch 2.0.1+cu118
• Datasets 1.18.3
• Tokenizers 0.13.3