ls-timit-100percent-supervised-meta

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0649
• Wer: 0.0253

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: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0964	7.04	1000	0.0706	0.0342
0.0445	14.08	2000	0.0649	0.0253

Framework versions

• Transformers 4.16.2
• Pytorch 1.10.2
• Datasets 1.18.2
• Tokenizers 0.10.3