whisper-large-v3-japanese-4k-steps

This model is a fine-tuned version of openai/whisper-large-v3 on the Common Voice 16.1 dataset. I followed a post by Sanchit Gandhi, https://huggingface.co/blog/fine-tune-whisper It took 24 hours using an A100 on Google Colab to complete 4000 steps using the Common Voice 16.1 dataset. Training loss dropped over epochs but validation loss increased, so textbook overfitting. Furthermore, WER increased. It achieves the following results on the evaluation set:

• Loss: 0.4057
• Wer: 18.2149

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• 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
0.1374	1.02	1000	0.3618	11.983182
0.0508	2.04	2000	0.3658	17.554657
0.0206	3.05	3000	0.3904	21.087484
0.0066	4.07	4000	0.4057	18.214909

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.17.0
• Tokenizers 0.15.2