whisper-large-v3-ft-btb-cv-cvad-ca-wlga-cy-2511

This model is a fine-tuned version of openai/whisper-large-v3 on the techiaith/banc-trawsgrifiadau-bangor train 25.10, techiaith/corpws-clllc-wlga clips main, techiaith/commonvoice_23_0_cy train+dev+other_with_excluded main, cymen-arfor/lleisiau-arfor train+dev main, techiaith/commonvoice_vad_cy train main dataset. It achieves the following results on the evaluation set:

• Loss: 0.3635
• Wer: 0.2736

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
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• training_steps: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5932	0.2523	500	0.5774	0.4085
0.491	0.5047	1000	0.4778	0.3715
0.4394	0.7570	1500	0.4366	0.3477
0.3786	1.0091	2000	0.4134	0.3097
0.3156	1.2614	2500	0.3972	0.3006
0.3094	1.5138	3000	0.3822	0.3114
0.2971	1.7661	3500	0.3710	0.2953
0.2295	2.0182	4000	0.3697	0.2828
0.2038	2.2705	4500	0.3679	0.2814
0.2049	2.5228	5000	0.3650	0.2791
0.2011	2.7752	5500	0.3635	0.2765
0.1793	3.0273	6000	0.3635	0.2731

Framework versions

• Transformers 4.52.4
• Pytorch 2.9.0+cu128
• Datasets 4.4.0
• Tokenizers 0.21.4