patrickvonplaten's picture
Update README.md
36bf72f
|
raw
history blame
2 kB
metadata
tags:
  - automatic-speech-recognition
  - librispeech_asr
  - generated_from_trainer
  - wavlm_libri_finetune
model-index:
  - name: wavlm-libri-clean-100h-base
    results: []

wavlm-libri-clean-100h-base

This model is a fine-tuned version of microsoft/wavlm-base on the LIBRISPEECH_ASR - CLEAN dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0829
  • Wer: 0.0675

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: 4
  • eval_batch_size: 4
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 8
  • total_train_batch_size: 32
  • total_eval_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: 3.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
2.8805 0.34 300 2.8686 1.0
0.2459 0.67 600 0.1858 0.1554
0.1114 1.01 900 0.1379 0.1191
0.0867 1.35 1200 0.1130 0.0961
0.0698 1.68 1500 0.1032 0.0877
0.0663 2.02 1800 0.0959 0.0785
0.0451 2.35 2100 0.0887 0.0748
0.0392 2.69 2400 0.0859 0.0698

Framework versions

  • Transformers 4.15.0.dev0
  • Pytorch 1.9.0+cu111
  • Datasets 1.16.2.dev0
  • Tokenizers 0.10.3