lutskivd/whisper-canadian-finetune-v2

Whisper tiny Canadian v2

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

• Loss: 0.4733
• Wer: 21.0595

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: 2
• eval_batch_size: 1
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1024
• training_steps: 4000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2777	0.5	500	0.4349	21.8988
0.3088	1.0	1000	0.4264	21.3480
0.1687	1.5	1500	0.4489	21.0595
0.134	2.0	2000	0.4428	20.5612
0.062	2.5	2500	0.4531	19.6171
0.0619	3.0	3000	0.4617	21.1907
0.0119	3.5	3500	0.4788	21.4267
0.0074	4.0	4000	0.4733	21.0595

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1