Edit model card

he-cantillation

This model is a fine-tuned version of ivrit-ai/whisper-v2-pd1-e1 on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.1078
  • Wer: 15.5291
  • Avg Precision Exact: 0.8760
  • Avg Recall Exact: 0.8758
  • Avg F1 Exact: 0.8753
  • Avg Precision Letter Shift: 0.8960
  • Avg Recall Letter Shift: 0.8959
  • Avg F1 Letter Shift: 0.8953
  • Avg Precision Word Level: 0.8987
  • Avg Recall Word Level: 0.8993
  • Avg F1 Word Level: 0.8984
  • Avg Precision Word Shift: 0.9450
  • Avg Recall Word Shift: 0.9466
  • Avg F1 Word Shift: 0.9451
  • Precision Median Exact: 0.9231
  • Recall Median Exact: 0.9231
  • F1 Median Exact: 0.9286
  • 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: 0.0769
  • Recall Min Word Shift: 0.0909
  • F1 Min Word Shift: 0.0833

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: 10
  • training_steps: 3000
  • 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
0.1518 0.0517 1000 0.1823 26.0329 0.7790 0.7842 0.7807 0.8094 0.8153 0.8114 0.8144 0.8207 0.8167 0.8953 0.9056 0.8994 0.8462 0.8571 0.8571 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
0.0993 0.1033 2000 0.1316 18.8804 0.8477 0.8540 0.8501 0.8691 0.8757 0.8716 0.8734 0.8799 0.8759 0.9264 0.9342 0.9294 0.9091 0.9167 0.9091 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0769 0.0909 0.0833
0.0625 0.1550 3000 0.1078 15.5291 0.8760 0.8758 0.8753 0.8960 0.8959 0.8953 0.8987 0.8993 0.8984 0.9450 0.9466 0.9451 0.9231 0.9231 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 0.0769 0.0909 0.0833

Framework versions

  • Transformers 4.41.2
  • Pytorch 2.2.1
  • Datasets 2.20.0
  • Tokenizers 0.19.1
Downloads last month
0
Safetensors
Model size
1.54B 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.

Model tree for cantillation/Teamim-large-v2_WeightDecay-0.05_Augmented_Combined-Data_date-09-07-2024_16-25

Finetuned
(6)
this model
Finetunes
1 model