Whisper small vi - Ox

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2023
  • Wer: 12.5320

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: 1.25e-05
  • train_batch_size: 8
  • eval_batch_size: 2
  • seed: 42
  • optimizer: Use 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: 500
  • num_epochs: 3.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.469 0.0156 1000 0.4203 24.3421
0.347 0.0311 2000 0.3729 22.5519
0.3173 0.0467 3000 0.3462 21.2276
0.3297 0.0623 4000 0.3402 19.7556
0.3225 0.0779 5000 0.3223 19.5226
0.2993 0.0934 6000 0.3068 17.3288
0.3226 0.1090 7000 0.2958 17.3288
0.2842 0.1246 8000 0.2881 16.8116
0.287 0.1401 9000 0.2847 16.3569
0.3101 0.1557 10000 0.2757 15.9477
0.2852 0.1713 11000 0.2680 15.7715
0.2617 0.1868 12000 0.2659 15.9136
0.2534 0.2024 13000 0.2596 15.5499
0.2388 0.2180 14000 0.2577 15.1520
0.227 0.2336 15000 0.2548 14.7258
0.2418 0.2491 16000 0.2515 14.5325
0.2521 0.2647 17000 0.2518 14.5212
0.2768 0.2803 18000 0.2473 14.3677
0.2552 0.2958 19000 0.2432 13.8221
0.2395 0.3114 20000 0.2370 14.6519
0.2189 0.3270 21000 0.2336 13.6232
0.2226 0.3426 22000 0.2314 13.2197
0.2273 0.3581 23000 0.2328 13.5607
0.2571 0.3737 24000 0.2287 13.0889
0.2629 0.3893 25000 0.2315 13.0946
0.2181 0.4048 26000 0.2267 12.8275
0.2191 0.4204 27000 0.2296 13.6516
0.2189 0.4360 28000 0.2248 13.0605
0.2471 0.4515 29000 0.2235 12.5718
0.2037 0.4671 30000 0.2189 12.7593
0.2077 0.4827 31000 0.2187 13.2822
0.1795 0.4983 32000 0.2188 12.7138
0.2084 0.5138 33000 0.2177 12.7707
0.2355 0.5294 34000 0.2083 12.0318
0.2105 0.5450 35000 0.2082 12.1512
0.2059 0.5605 36000 0.2122 12.3330
0.1918 0.5761 37000 0.2076 11.8045
0.1812 0.5917 38000 0.2068 11.9636
0.1909 0.6072 39000 0.2083 13.4527
0.2454 0.6228 40000 0.2075 11.8272
0.2087 0.6384 41000 0.2031 11.6965
0.1988 0.6540 42000 0.2026 11.5374
0.1917 0.6695 43000 0.2025 11.5203
0.1688 0.6851 44000 0.2023 12.5320

Framework versions

  • Transformers 4.48.3
  • Pytorch 2.6.0+cu124
  • Datasets 3.2.0
  • Tokenizers 0.21.0
Downloads last month
19
Safetensors
Model size
242M params
Tensor type
F32
·
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.

Model tree for aleni/whisper-vismall

Finetuned
(2395)
this model