README.md

metadata

library_name: transformers
language:
  - ru
license: mit
base_model: openai/whisper-large-v3-turbo
tags:
  - generated_from_trainer
datasets:
  - skinnynpale/sasha-ai-asr-dataset-22
metrics:
  - wer
model-index:
  - name: Whisper Large V3 Turbo Russian
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: skinnynpale/sasha-ai-asr-dataset-22
          type: skinnynpale/sasha-ai-asr-dataset-22
          args: 'split: train+validation'
        metrics:
          - name: Wer
            type: wer
            value: 0.4558936614302977

Whisper Large V3 Turbo Russian

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the skinnynpale/sasha-ai-asr-dataset-22 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0165
• Wer: 0.4559

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: 8
• 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: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0831	0.8183	1000	0.0442	1.6258
0.0973	1.6367	2000	0.0251	1.1095
0.077	2.4550	3000	0.0204	0.7415
0.0515	3.2733	4000	0.0165	0.4559

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1