CoolFace
Modelpublic

cantillation/Teamim-distil-large-v3.5_WeightDecay-0.005_Augmented__WithSRT_LR-1e-05_date-10-06-2025

sourceHugging Facemitupdated 1y agoView on Hugging Face
0likes26downloads
Model Card

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

he-cantillation

This model is a fine-tuned version of distil-whisper/distil-large-v3.5 on an unknown dataset. It achieves the following results on the evaluation set:

  • —Loss: 0.7945
  • —Wer: 68.9512
  • —Avg Precision Exact: 0.2622
  • —Avg Recall Exact: 0.2502
  • —Avg F1 Exact: 0.2513
  • —Avg Precision Letter Shift: 0.2922
  • —Avg Recall Letter Shift: 0.2803
  • —Avg F1 Letter Shift: 0.2796
  • —Avg Precision Word Level: 0.3129
  • —Avg Recall Word Level: 0.3458
  • —Avg F1 Word Level: 0.3227
  • —Avg Precision Word Shift: 0.5277
  • —Avg Recall Word Shift: 0.5512
  • —Avg F1 Word Shift: 0.5260
  • —Precision Median Exact: 0.1429
  • —Recall Median Exact: 0.1379
  • —F1 Median Exact: 0.1333
  • —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: 0.0
  • —Recall Min Word Shift: 0.0
  • —F1 Min Word Shift: 0.0

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: 1e-05
  • —trainbatchsize: 16
  • —evalbatchsize: 2
  • —seed: 42
  • —optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
  • —lrschedulertype: linear
  • —lrschedulerwarmup_steps: 100
  • —training_steps: 10000
  • —mixedprecisiontraining: Native AMP

Training results

Training LossEpochStepValidation LossWerAvg Precision ExactAvg Recall ExactAvg F1 ExactAvg Precision Letter ShiftAvg Recall Letter ShiftAvg F1 Letter ShiftAvg Precision Word LevelAvg Recall Word LevelAvg F1 Word LevelAvg Precision Word ShiftAvg Recall Word ShiftAvg F1 Word ShiftPrecision Median ExactRecall Median ExactF1 Median ExactPrecision Max ExactRecall Max ExactF1 Max ExactPrecision Min ExactRecall Min ExactF1 Min ExactPrecision Min Letter ShiftRecall Min Letter ShiftF1 Min Letter ShiftPrecision Min Word LevelRecall Min Word LevelF1 Min Word LevelPrecision Min Word ShiftRecall Min Word ShiftF1 Min Word Shift
No log0.0001112.1921169.30430.00.00.00.00.00.00.00.00.00.00.00.00.00.00.0000000000000000
0.0960.592550000.950376.89180.19740.20420.19810.23020.24050.23120.25190.28060.26070.46840.51930.48280.10710.11110.10531.01.01.00.00.00.00.00.00.00.00.00.00.00.00.0
0.06141.1850100000.794568.95120.26220.25020.25130.29220.28030.27960.31290.34580.32270.52770.55120.52600.14290.13790.13331.01.01.00.00.00.00.00.00.00.00.00.00.00.00.0

Framework versions

  • —Transformers 4.49.0
  • —Pytorch 2.7.0+cu126
  • —Datasets 2.12.0
  • —Tokenizers 0.20.1