Edit model card

Whisper Small Hu CV17 (ez a modell csak tesztelésre készült, a használata hibákat okozhat)

This model is a fine-tuned version of openai/whisper-small on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0862
  • Wer Ortho: 6.536794
  • Wer: 5.627038

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: 2.5e-05
  • train_batch_size: 64
  • 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: 500
  • training_steps: 6000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Ortho Wer
0.3208 0.3298 250 0.3558 34.5839 31.0969
0.2542 0.6596 500 0.2683 27.2010 24.2218
0.2204 0.9894 750 0.2035 22.2962 19.4960
0.1301 1.3193 1000 0.1673 17.6683 15.2446
0.1171 1.6491 1250 0.1433 15.2445 12.9677
0.1075 1.9789 1500 0.1192 12.8658 10.7827
0.0492 2.3087 1750 0.1091 11.2489 9.1877
0.0491 2.6385 2000 0.1046 10.7582 9.0543
0.0472 2.9683 2250 0.0945 9.0750 7.5037
0.0203 3.2982 2500 0.0932 8.5873 7.2428
0.0188 3.6280 2750 0.0900 8.2019 6.9671
0.0183 3.9578 3000 0.0859 7.5154 6.4008
0.0068 4.2876 3250 0.0837 7.2052 6.1785
0.0064 4.6174 3500 0.0843 6.9613 6.0213
0.0061 4.9472 3750 0.0847 6.9674 5.9383
0.0025 5.2770 4000 0.0851 6.6994 5.7723
0.0023 5.6069 4250 0.0847 6.6331 5.6863
0.0022 5.9367 4500 0.0844 6.6211 5.7338
0.0014 6.2665 4750 0.0855 6.5789 5.6834
0.0011 6.5963 5000 0.0856 6.5609 5.6567
0.0012 6.9261 5250 0.0862 6.5368 5.6270
0.0009 7.2559 5500 0.0870 6.5699 5.6597
0.0009 7.5858 5750 0.0873 6.5639 5.6597
0.001 7.9156 6000 0.0873 6.5518 5.6389

Framework versions

  • Transformers 4.41.0
  • Pytorch 2.2.1+cu121
  • Datasets 2.19.1
  • Tokenizers 0.19.1
Downloads last month
2
Safetensors
Model size
242M params
Tensor type
F32
·

Finetuned from

Evaluation results