wav2vec2-base-mn-pretrain-42h-mn-silence-speech-commands

This model is a fine-tuned version of bayartsogt/wav2vec2-base-mn-pretrain-42h on the Mongolian Speech Commands dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0562
  • Mn Acc: 0.9830
  • Mn F1: 0.9832
  • Silence Acc: 1.0
  • Silence F1: 1.0

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: 5e-05
  • train_batch_size: 128
  • eval_batch_size: 128
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 8

Training results

Training Loss Epoch Step Validation Loss Mn Acc Mn F1 Silence Acc Silence F1
No log 0.4 8 2.0276 0.0455 0.0239 1.0 1.0
2.3615 0.8 16 1.1112 0.0057 0.0108 1.0 1.0
2.0154 1.2 24 0.6836 0.6307 0.5627 0.9975 0.9988
1.5733 1.6 32 0.4493 0.7898 0.7652 0.9975 0.9988
1.1148 2.0 40 0.3264 0.8409 0.8202 1.0 1.0
1.1148 2.4 48 0.2490 0.8864 0.8768 1.0 1.0
0.7937 2.8 56 0.1739 0.9545 0.9540 1.0 1.0
0.586 3.2 64 0.1425 0.9659 0.9664 1.0 1.0
0.4445 3.6 72 0.1137 0.9659 0.9659 1.0 1.0
0.3892 4.0 80 0.0942 0.9773 0.9772 1.0 1.0
0.3892 4.4 88 0.0914 0.9716 0.9717 1.0 1.0
0.3341 4.8 96 0.0748 0.9773 0.9775 1.0 1.0
0.2863 5.2 104 0.0670 0.9886 0.9886 1.0 1.0
0.2622 5.6 112 0.0697 0.9830 0.9832 1.0 1.0
0.2222 6.0 120 0.0638 0.9830 0.9832 1.0 1.0
0.2222 6.4 128 0.0580 0.9886 0.9886 1.0 1.0
0.213 6.8 136 0.0575 0.9830 0.9832 1.0 1.0
0.2082 7.2 144 0.0587 0.9830 0.9832 1.0 1.0
0.202 7.6 152 0.0582 0.9830 0.9832 1.0 1.0
0.1936 8.0 160 0.0562 0.9830 0.9832 1.0 1.0

Framework versions

  • Transformers 4.30.2
  • Pytorch 2.0.0
  • Datasets 2.14.4
  • Tokenizers 0.13.3
Downloads last month
20
Safetensors
Model size
94.6M 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.