Edit model card

wav2vec2-bert-CV16-en-libri

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the LIBRISPEECH_ASR - CLEAN dataset. It achieves the following results on the evaluation set:

  • Loss: 0.1035
  • Wer: 0.0708
  • Cer: 0.0194

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: 3e-05
  • train_batch_size: 12
  • eval_batch_size: 12
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 3
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 72
  • total_eval_batch_size: 36
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 10000
  • num_epochs: 7.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Cer Validation Loss Wer
2.8812 0.63 250 1.0000 2.8923 1.0
1.2899 1.26 500 0.2563 1.1471 0.7030
0.5276 1.89 750 0.1127 0.4687 0.4114
0.3313 2.52 1000 0.0659 0.2870 0.2577
0.2089 3.16 1250 0.0445 0.2079 0.1766
0.1634 3.79 1500 0.0366 0.1687 0.1411
0.1546 4.42 1750 0.1452 0.1138 0.0294
0.1245 5.05 2000 0.1316 0.0973 0.0260
0.1341 5.68 2250 0.1196 0.0867 0.0234
0.0942 6.31 2500 0.1128 0.0794 0.0213
0.0848 6.94 2750 0.1077 0.0717 0.0197

Framework versions

  • Transformers 4.37.0.dev0
  • Pytorch 2.1.0+cu121
  • Datasets 2.16.1
  • Tokenizers 0.15.0
Downloads last month
18
Safetensors
Model size
606M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for ylacombe/wav2vec2-bert-CV16-en-libri

Finetuned
(169)
this model
Finetunes
1 model