Edit model card

wav2vec2-large-xls-r-300m-gn-pt

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

  • Loss: 0.6822
  • Wer: 0.5432

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: 16
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 100
  • num_epochs: 35
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
4.1972 0.79 400 1.9288 1.0045
0.9928 1.58 800 0.8247 0.9452
0.6075 2.36 1200 0.7675 0.8451
0.4724 3.15 1600 0.5485 0.7111
0.3879 3.94 2000 0.5885 0.7433
0.3152 4.73 2400 0.7606 0.7695
0.2872 5.52 2800 0.5723 0.6608
0.258 6.31 3200 0.5971 0.6820
0.2317 7.09 3600 0.5845 0.6471
0.2137 7.88 4000 0.7690 0.7198
0.193 8.67 4400 0.6219 0.6614
0.1795 9.46 4800 0.6203 0.6703
0.1768 10.25 5200 0.5645 0.6164
0.1637 11.03 5600 0.5804 0.6412
0.1573 11.82 6000 0.5914 0.5896
0.1467 12.61 6400 0.6517 0.6200
0.141 13.4 6800 0.6376 0.6310
0.135 14.19 7200 0.6343 0.6042
0.1279 14.98 7600 0.6680 0.6325
0.1222 15.76 8000 0.7109 0.6617
0.1169 16.55 8400 0.7067 0.6361
0.114 17.34 8800 0.7143 0.6144
0.1085 18.13 9200 0.6871 0.6081
0.0996 18.92 9600 0.8332 0.6569
0.0952 19.7 10000 0.7076 0.5992
0.0929 20.49 10400 0.6946 0.6078
0.0871 21.28 10800 0.6197 0.5822
0.0823 22.07 11200 0.6969 0.5876
0.0776 22.86 11600 0.6285 0.5619
0.0758 23.65 12000 0.7098 0.6010
0.0728 24.43 12400 0.6618 0.5905
0.0664 25.22 12800 0.6484 0.5536
0.0656 26.01 13200 0.6417 0.5593
0.0603 26.8 13600 0.7287 0.5813
0.0571 27.59 14000 0.6727 0.5700
0.0559 28.37 14400 0.6775 0.5631
0.0555 29.16 14800 0.7849 0.5968
0.0506 29.95 15200 0.8266 0.6185
0.0485 30.74 15600 0.7347 0.5747
0.0461 31.53 16000 0.6836 0.5432
0.0423 32.32 16400 0.6913 0.5396
0.0407 33.1 16800 0.6655 0.5328
0.04 33.89 17200 0.6873 0.5399
0.0396 34.68 17600 0.6822 0.5432

Framework versions

  • Transformers 4.35.0
  • Pytorch 2.0.1+cu118
  • Datasets 2.14.6
  • Tokenizers 0.14.1
Downloads last month
1
Safetensors
Model size
315M params
Tensor type
F32
·

Finetuned from

Evaluation results