whisper-new-nnat-20h-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: 0.4515
• Wer: 20.4140

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: 104
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5877	0.2989	104	0.5724	26.2118
0.3882	0.5977	208	0.5024	22.1608
0.3917	0.8966	312	0.4835	21.9548
0.3331	1.1954	416	0.4741	21.4818
0.3337	1.4943	520	0.4647	22.0812
0.3604	1.7931	624	0.4574	21.0088
0.3182	2.0920	728	0.4540	21.0134
0.3104	2.3908	832	0.4552	20.5685
0.2996	2.6897	936	0.4529	20.3718
0.2894	2.9885	1040	0.4515	20.4140

Framework versions

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