xls-r-1b-bem-natbed-non-native-model

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the NATBED - BEM dataset. It achieves the following results on the evaluation set:

• Loss: 0.6818
• Wer: 0.7142

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.8975	0.4515	100	1.1343	0.9316
1.0247	0.9029	200	0.9811	0.8826
0.9631	1.3544	300	0.8108	0.7524
0.8711	1.8059	400	0.7749	0.7638
0.7935	2.2573	500	0.8008	0.8232
0.7869	2.7088	600	0.7334	0.7262
0.7265	3.1603	700	0.7239	0.6950
0.7011	3.6117	800	0.7077	0.6807
0.7343	4.0632	900	0.6985	0.6914
0.6624	4.5147	1000	0.6818	0.7142
0.6642	4.9661	1100	0.7097	0.6968
0.6064	5.4176	1200	0.7195	0.6856
0.5867	5.8691	1300	0.6899	0.6880

Framework versions

• Transformers 4.46.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0