Edit model card

w2v2-ks-jpqd-quant-all-finetuned-student

This model is a fine-tuned version of anton-l/wav2vec2-base-ft-keyword-spotting on the superb dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0933
  • Accuracy: 0.9769

This model is quantized. The input is also quantized. Structured Sparsity in transformer block linear layers is 64%.

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: 7e-05
  • train_batch_size: 32
  • eval_batch_size: 32
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 128
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.5
  • num_epochs: 12.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Accuracy
0.4481 1.0 399 0.2105 0.9469
5.6584 2.0 798 5.5480 0.9428
8.7915 3.0 1197 8.6634 0.9601
10.4775 4.0 1596 10.2819 0.9553
10.9142 5.0 1995 10.7770 0.9657
10.9478 6.0 2394 10.7637 0.9660
0.2765 7.0 2793 0.1335 0.9678
0.2532 8.0 3192 0.1075 0.9732
0.2837 9.0 3591 0.1109 0.9700
0.2 10.0 3990 0.1006 0.9765
0.1742 11.0 4389 0.0930 0.9776
0.1718 12.0 4788 0.0933 0.9769

Framework versions

  • Transformers 4.26.0
  • Pytorch 1.13.1+cu116
  • Datasets 2.8.0
  • Tokenizers 0.13.2
Downloads last month
3

Dataset used to train yujiepan/internal.wav2vec2-base-superb-ks-int8-structured64-quantize-inputs