metadata
language:
- ru
license: apache-2.0
tags:
- whisper-event
- generated_from_trainer
datasets:
- mozilla-foundation/common_voice_11_0
metrics:
- wer
model-index:
- name: Whisper Medium Russian
results:
- task:
type: automatic-speech-recognition
name: Automatic Speech Recognition
dataset:
name: mozilla-foundation/common_voice_11_0 ru
type: mozilla-foundation/common_voice_11_0
config: ru
split: test
args: ru
metrics:
- type: wer
value: 7.562437929892964
name: Wer
- task:
type: automatic-speech-recognition
name: Automatic Speech Recognition
dataset:
name: google/fleurs
type: google/fleurs
config: ru_ru
split: test
metrics:
- type: wer
value: 10.92
name: WER
Whisper Medium Russian
This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_11_0 ru dataset. It achieves the following results on the evaluation set:
- Loss: 0.2253
- Wer: 7.5624
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: 16
- eval_batch_size: 16
- 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: 10000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.1578 | 0.1 | 1000 | 0.1662 | 8.8290 |
| 0.045 | 1.08 | 2000 | 0.1748 | 8.9148 |
| 0.0176 | 2.06 | 3000 | 0.1889 | 8.7848 |
| 0.0104 | 3.04 | 4000 | 0.1922 | 8.4354 |
| 0.0051 | 4.02 | 5000 | 0.2034 | 8.1865 |
| 0.0047 | 4.12 | 6000 | 0.2012 | 8.0455 |
| 0.0018 | 5.1 | 7000 | 0.2117 | 7.6237 |
| 0.0004 | 6.08 | 8000 | 0.2177 | 7.6078 |
| 0.0003 | 7.06 | 9000 | 0.2244 | 7.6262 |
| 0.0002 | 8.04 | 10000 | 0.2253 | 7.5624 |
Framework versions
- Transformers 4.28.0.dev0
- Pytorch 2.0.0+cu117
- Datasets 2.11.1.dev0
- Tokenizers 0.13.2