whisper-new-nnat-10h-caps-ecapa

This model is a fine-tuned version of openai/whisper-large-v2 on the JASMIN-CGN dataset. It achieves the following results on the evaluation set:

• Loss: 0.4781
• Wer: 21.6925

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: 0.0001
• train_batch_size: 48
• eval_batch_size: 32
• seed: 42
• 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: linear
• lr_scheduler_warmup_steps: 59
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4332	0.2995	59	0.6828	31.1806
0.4937	0.5990	118	0.5383	23.9873
0.3968	0.8985	177	0.5136	23.0366
0.402	1.1980	236	0.5016	22.7369
0.3518	1.4975	295	0.4921	21.8517
0.3556	1.7970	354	0.4859	22.0344
0.3345	2.0964	413	0.4839	21.9313
0.3128	2.3959	472	0.4810	21.9220
0.3074	2.6954	531	0.4791	21.7253
0.3198	2.9949	590	0.4781	21.6925

Framework versions

• PEFT 0.17.1
• Transformers 4.57.6
• Pytorch 2.8.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2