Hungarians/whisper-tiny-cv16-hu-v3

Whisper Tiny Hu v3

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 16.0 dataset. It achieves the following results on the evaluation set:

• eval_loss: 0.2041
• eval_wer_ortho: 13.8474
• eval_wer: 12.7928
• eval_runtime: 1002.1306
• eval_samples_per_second: 4.621
• eval_steps_per_second: 0.578
• epoch: 13.39
• step: 20000

Model description

Hungarian language trained modell. Faster-whisper versions in int8, fp16 and ft32 floders.

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: 4e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 500
• training_steps: 20000
• mixed_precision_training: Native AMP

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0