Whisper Medium Dv - Leon Lee

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 13 dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2803
  • Wer Ortho: 48.8335
  • Wer: 8.4327

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: 4
  • eval_batch_size: 4
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 8
  • optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 100
  • training_steps: 8000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Ortho Wer
0.1344 0.8157 500 0.1613 59.9206 12.1049
0.0732 1.6313 1000 0.1382 52.9285 10.2271
0.0411 2.4470 1500 0.1447 52.3087 9.7628
0.0244 3.2626 2000 0.1538 51.6749 9.4534
0.0164 4.0783 2500 0.1839 53.8617 9.4290
0.0162 4.8940 3000 0.1734 51.7863 9.0604
0.0086 5.7096 3500 0.1962 50.8949 9.0222
0.0048 6.5253 4000 0.2299 50.7904 8.8205
0.003 7.3409 4500 0.2336 50.7487 8.8344
0.0017 8.1566 5000 0.2303 50.2472 8.6275
0.0017 8.9723 5500 0.2455 49.9896 8.6327
0.0005 9.7879 6000 0.2551 49.8015 8.5371
0.0001 10.6036 6500 0.2682 48.8962 8.4414
0.0 11.4192 7000 0.2732 48.6663 8.4206
0.0 12.2349 7500 0.2800 48.8892 8.4605
0.0 13.0506 8000 0.2803 48.8335 8.4327

Framework versions

  • Transformers 4.48.1
  • Pytorch 2.6.0+cu124
  • Datasets 3.2.0
  • Tokenizers 0.21.0
Downloads last month
3
Safetensors
Model size
764M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for LeonM78Code/whisper-medium-dv

Finetuned
(697)
this model

Dataset used to train LeonM78Code/whisper-medium-dv

Evaluation results