xls-r-1b-bem-natbed-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: 0.6841
• Wer: 0.7487

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
4.5137	0.5618	100	2.5549	1.0
1.3916	1.1236	200	1.0883	0.9840
0.9962	1.6854	300	0.8153	0.8190
0.8625	2.2472	400	0.8690	0.8418
0.8168	2.8090	500	0.7395	0.7390
0.7197	3.3708	600	0.7596	0.7366
0.6848	3.9326	700	0.7033	0.7229
0.6134	4.4944	800	0.8300	0.7662
0.6303	5.0562	900	0.7365	0.7896
0.5467	5.6180	1000	0.6841	0.7487
0.5194	6.1798	1100	0.7868	0.6949
0.4617	6.7416	1200	0.7563	0.7278
0.4525	7.3034	1300	0.7276	0.6730

Framework versions

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