whisper-large-v3-ft-cv-cy-train-all

This model is a fine-tuned version of openai/whisper-large-v3 on the DewiBrynJones/commonvoice_18_0_cy default dataset. It achieves the following results on the evaluation set:

• Loss: 0.3638
• Wer: 0.1817

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1429	1.9455	1000	0.2754	0.2208
0.0232	3.8911	2000	0.2916	0.1991
0.0046	5.8366	3000	0.3219	0.1878
0.0009	7.7821	4000	0.3454	0.1832
0.0004	9.7276	5000	0.3638	0.1817

Framework versions

• Transformers 4.44.0
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1