whisper-large-v2-ft-cy-2603

This model is a fine-tuned version of openai/whisper-large-v2 on the DewiBrynJones/preprocessed-whisper-btb-cv-cvad-wlga-ca-2603 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3972
• Wer: 0.2766

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_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: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5809	0.2037	500	0.5988	0.4648
0.4833	0.4073	1000	0.5038	0.4001
0.4539	0.6110	1500	0.4565	0.3717
0.4401	0.8147	2000	0.4278	0.3507
0.3308	1.0183	2500	0.4103	0.3679
0.309	1.2220	3000	0.4017	0.3542
0.3151	1.4257	3500	0.3900	0.3468
0.3113	1.6293	4000	0.3850	0.3032
0.304	1.8330	4500	0.3773	0.3499
0.2092	2.0367	5000	0.3820	0.2956
0.2196	2.2403	5500	0.3861	0.2991
0.2107	2.4440	6000	0.3766	0.2923
0.2175	2.6477	6500	0.3737	0.2908
0.2131	2.8513	7000	0.3704	0.2947
0.1318	3.0550	7500	0.4003	0.2861
0.1336	3.2587	8000	0.4034	0.2873
0.1357	3.4623	8500	0.3964	0.2796
0.1359	3.6660	9000	0.3972	0.2766
0.1352	3.8697	9500	0.3973	0.2870
0.0778	4.0733	10000	0.4422	0.2797
0.0724	4.2770	10500	0.4552	0.2826
0.0749	4.4807	11000	0.4551	0.2807
0.0746	4.6843	11500	0.4582	0.2807
0.071	4.8880	12000	0.4608	0.2810
0.0468	5.0916	12500	0.5011	0.2844
0.0462	5.2953	13000	0.5058	0.2854
0.0437	5.4990	13500	0.5132	0.2849
0.046	5.7026	14000	0.5117	0.2829

Framework versions

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