wavlm-large-timit-punctuation

This model is a fine-tuned version of microsoft/wavlm-large on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3368
  • Wer: 0.2601

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: 8
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 1000
  • num_epochs: 30
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
5.2379 1.0 500 3.1228 1.0
2.5847 2.01 1000 1.1550 0.9147
1.0034 3.01 1500 0.5856 0.5180
0.5868 4.02 2000 0.4238 0.4229
0.3892 5.02 2500 0.3356 0.3665
0.2926 6.02 3000 0.3196 0.3360
0.2294 7.03 3500 0.3046 0.3170
0.1976 8.03 4000 0.3032 0.3111
0.1644 9.04 4500 0.2946 0.2954
0.1574 10.04 5000 0.3211 0.2998
0.1391 11.04 5500 0.2986 0.2922
0.1124 12.05 6000 0.2948 0.2837
0.1003 13.05 6500 0.2928 0.2788
0.1031 14.06 7000 0.3230 0.2805
0.0901 15.06 7500 0.3081 0.2749
0.0842 16.06 8000 0.3075 0.2726
0.0809 17.07 8500 0.3215 0.2717
0.0747 18.07 9000 0.3272 0.2721
0.0735 19.08 9500 0.3242 0.2684
0.0631 20.08 10000 0.3216 0.2640
0.0632 21.08 10500 0.3149 0.2646
0.0625 22.09 11000 0.3196 0.2630
0.0611 23.09 11500 0.3244 0.2638
0.0532 24.1 12000 0.3271 0.2641
0.0503 25.1 12500 0.3368 0.2636
0.0534 26.1 13000 0.3393 0.2627
0.049 27.11 13500 0.3389 0.2626
0.0441 28.11 14000 0.3375 0.2605
0.0522 29.12 14500 0.3368 0.2601

Framework versions

  • Transformers 4.19.2
  • Pytorch 1.8.2+cu111
  • Datasets 1.17.0
  • Tokenizers 0.11.6
Downloads last month
26
Safetensors
Model size
316M 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.