Edit model card

wav2vec2-large-xlsr-53-thai

This model is a fine-tuned version of airesearch/wav2vec2-large-xlsr-53-th on the common_voice dataset. It achieves the following results on the evaluation set:

  • Loss: 3.3576
  • Wer: 0.7431

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

Training results

Training Loss Epoch Step Validation Loss Wer
6.7312 3.33 100 3.3592 1.0
3.3687 6.67 200 3.2175 1.0
2.4527 10.0 300 2.2648 0.7911
1.0505 13.33 400 2.2322 0.7659
0.7725 16.67 500 2.2775 0.7505
0.6289 20.0 600 2.3209 0.7498
0.543 23.33 700 2.4494 0.7572
0.4991 26.67 800 2.5798 0.7597
0.4492 30.0 900 2.5685 0.7461
0.3737 33.33 1000 2.6186 0.7486
0.3358 36.67 1100 2.7781 0.7480
0.3247 40.0 1200 2.8999 0.7535
0.2963 43.33 1300 2.8668 0.7388
0.2825 46.67 1400 2.8983 0.7449
0.2651 50.0 1500 2.9699 0.7461
0.2597 53.33 1600 2.9930 0.7314
0.2629 56.67 1700 2.9852 0.7406
0.2406 60.0 1800 3.0552 0.7474
0.2293 63.33 1900 3.1058 0.7344
0.2193 66.67 2000 3.1594 0.7406
0.2174 70.0 2100 3.2351 0.7369
0.2127 73.33 2200 3.2696 0.7388
0.2061 76.67 2300 3.2954 0.7566
0.1947 80.0 2400 3.2878 0.7529
0.199 83.33 2500 3.3233 0.7486
0.1961 86.67 2600 3.3136 0.7437
0.1928 90.0 2700 3.3240 0.7406
0.1875 93.33 2800 3.3479 0.7425
0.1852 96.67 2900 3.3681 0.7425
0.1814 100.0 3000 3.3576 0.7431

Framework versions

  • Transformers 4.28.0
  • Pytorch 2.0.1+cu118
  • Datasets 1.16.1
  • Tokenizers 0.13.3
Downloads last month
5
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.

Dataset used to train BALAKA/wav2vec2-large-xlsr-53-thai

Evaluation results