he-cantillation

This model is a fine-tuned version of cantillation/Teamim-AllNusah-whisper-medium_Random-True_Mid on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.1410
  • Wer: 11.6348
  • Avg Precision Exact: 0.8905
  • Avg Recall Exact: 0.8961
  • Avg F1 Exact: 0.8928
  • Avg Precision Letter Shift: 0.9069
  • Avg Recall Letter Shift: 0.9133
  • Avg F1 Letter Shift: 0.9096
  • Avg Precision Word Level: 0.9091
  • Avg Recall Word Level: 0.9146
  • Avg F1 Word Level: 0.9114
  • Avg Precision Word Shift: 0.9658
  • Avg Recall Word Shift: 0.9719
  • Avg F1 Word Shift: 0.9683
  • Precision Median Exact: 0.9375
  • Recall Median Exact: 0.9412
  • F1 Median Exact: 0.9630
  • Precision Max Exact: 1.0
  • Recall Max Exact: 1.0
  • F1 Max Exact: 1.0
  • Precision Min Exact: 0.0
  • Recall Min Exact: 0.0
  • F1 Min Exact: 0.0
  • Precision Min Letter Shift: 0.0
  • Recall Min Letter Shift: 0.0
  • F1 Min Letter Shift: 0.0
  • Precision Min Word Level: 0.0
  • Recall Min Word Level: 0.0
  • F1 Min Word Level: 0.0
  • Precision Min Word Shift: 0.6
  • Recall Min Word Shift: 0.6429
  • F1 Min Word Shift: 0.6429

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: 1e-05
  • train_batch_size: 8
  • eval_batch_size: 32
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 100
  • training_steps: 5000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Avg Precision Exact Avg Recall Exact Avg F1 Exact Avg Precision Letter Shift Avg Recall Letter Shift Avg F1 Letter Shift Avg Precision Word Level Avg Recall Word Level Avg F1 Word Level Avg Precision Word Shift Avg Recall Word Shift Avg F1 Word Shift Precision Median Exact Recall Median Exact F1 Median Exact Precision Max Exact Recall Max Exact F1 Max Exact Precision Min Exact Recall Min Exact F1 Min Exact Precision Min Letter Shift Recall Min Letter Shift F1 Min Letter Shift Precision Min Word Level Recall Min Word Level F1 Min Word Level Precision Min Word Shift Recall Min Word Shift F1 Min Word Shift
No log 0.0 1 2.8792 88.9597 0.1569 0.1670 0.1602 0.1946 0.2074 0.1987 0.2144 0.2256 0.2176 0.3720 0.3988 0.3809 0.1088 0.125 0.1213 0.75 0.7692 0.7407 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
0.2893 0.04 500 0.1747 17.9830 0.8373 0.8390 0.8373 0.8615 0.8649 0.8624 0.8683 0.8713 0.8691 0.9388 0.9424 0.9398 0.9199 0.9199 0.9016 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
0.209 0.08 1000 0.1427 14.5648 0.8556 0.8629 0.8586 0.8786 0.8862 0.8818 0.8835 0.8907 0.8865 0.9515 0.9579 0.9541 0.9286 0.9286 0.9286 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.3333 0.3077 0.32
0.1831 0.12 1500 0.1381 14.2675 0.8740 0.8825 0.8776 0.8951 0.9045 0.8991 0.8990 0.9074 0.9026 0.9536 0.9605 0.9565 0.9286 0.9310 0.9333 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.6429 0.6207
0.1371 0.16 2000 0.1376 13.5244 0.8738 0.8789 0.8757 0.8951 0.9004 0.8971 0.8987 0.9032 0.9004 0.9588 0.9627 0.9602 0.9333 0.9333 0.9333 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6923 0.7143 0.7333
0.1138 0.2 2500 0.1359 12.7601 0.8774 0.8859 0.8811 0.8963 0.9055 0.9003 0.9003 0.9072 0.9032 0.9582 0.9665 0.9618 0.9333 0.9333 0.9600 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.6923 0.6429
0.1104 0.24 3000 0.1356 12.8450 0.8749 0.8821 0.8780 0.8912 0.8993 0.8947 0.8940 0.9010 0.8970 0.9582 0.9657 0.9614 0.9333 0.9333 0.9565 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.7273 0.6667
0.0909 0.28 3500 0.1409 12.4204 0.8808 0.8873 0.8835 0.9007 0.9080 0.9038 0.9031 0.9100 0.9060 0.9639 0.9706 0.9667 0.9333 0.9333 0.9488 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.6923 0.6429
0.09 0.32 4000 0.1370 12.0170 0.8886 0.8910 0.8893 0.9053 0.9085 0.9064 0.9079 0.9106 0.9088 0.9655 0.9685 0.9665 0.9375 0.9393 0.9630 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6667 0.7333 0.7143
0.0685 0.36 4500 0.1405 11.9533 0.8912 0.8946 0.8924 0.9079 0.9121 0.9095 0.9103 0.9140 0.9117 0.9650 0.9703 0.9672 0.9412 0.9412 0.9630 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.6429 0.6429
0.0647 0.4 5000 0.1410 11.6348 0.8905 0.8961 0.8928 0.9069 0.9133 0.9096 0.9091 0.9146 0.9114 0.9658 0.9719 0.9683 0.9375 0.9412 0.9630 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.6429 0.6429

Framework versions

  • Transformers 4.39.0.dev0
  • Pytorch 2.2.1+cu121
  • Datasets 2.16.1
  • Tokenizers 0.15.0
Downloads last month
6
Safetensors
Model size
764M 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.