O'zbekcha SpeechToText 5-versiyasi

Bu model facebook/wav2vec2-base va MOZILLA-FOUNDATION/COMMON_VOICE_10_0 - versiyasining dataseti bilan o'qitildi. Model o'qitilgandan keyin quyidagi natijalarga erishildi:

  • Xatolik: 1.8085
  • So'zlarning xatolik darajasi: 0.9421

Model haqida

Model 2 kun davomida 2xRTX3090 24GBli Video kartada o'qitildi.

Modelni o'qitish uchun quyidagi giperparameterlarni qo'ydik:

Quyidagi giperparameterlar model o'qitish jarayonida ishlatildi:

  • learning_rate: 3e-05
  • train_batch_size: 32
  • eval_batch_size: 16
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 2
  • total_train_batch_size: 64
  • total_eval_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 30.0
  • mixed_precision_training: Native AMP

O'qitish natijalari

Xatolik Epox Qadam Tasdiq xatoligi SXD
0.3452 5.45 5000 0.3839 0.4574
0.2466 10.91 10000 0.4011 0.4067
1.5753 16.36 15000 1.2937 0.8844
1.9454 21.81 20000 1.8227 0.9392
1.922 27.26 25000 1.8085 0.9421
Downloads last month
9
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support