whisper-large-v2-ft-cy-2601

This model is a fine-tuned version of openai/whisper-large-v2 on the DewiBrynJones/preprocessed-whisper-btb-cv-cvad-cven-wlga-ca-ec-2601 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3660
• Wer: 0.3715

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: 5e-06
• 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_fused 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: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9408	0.1966	500	0.8202	0.5409
0.5988	0.3932	1000	0.5391	0.5537
0.5247	0.5899	1500	0.4744	0.4518
0.4995	0.7865	2000	0.4416	0.4371
0.456	0.9831	2500	0.4165	0.4022
0.3781	1.1797	3000	0.4080	0.3901
0.3756	1.3763	3500	0.3982	0.4062
0.3763	1.5729	4000	0.3892	0.3990
0.3509	1.7696	4500	0.3777	0.3831
0.3599	1.9662	5000	0.3745	0.3888
0.2981	2.1628	5500	0.3758	0.3634
0.29	2.3594	6000	0.3706	0.3904
0.2829	2.5560	6500	0.3677	0.3634
0.2892	2.7527	7000	0.3660	0.3789
0.286	2.9493	7500	0.3657	0.3772
0.2641	3.1459	8000	0.3660	0.3715

Framework versions

• Transformers 4.57.6
• Pytorch 2.10.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2