whisper-new-nnat-5h-mmr-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: 0.5545
• Wer: 35.1051

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: 62
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9171	0.2967	62	0.7342	44.3497
0.377	0.5933	124	0.6011	41.0902
0.3449	0.8900	186	0.5843	42.5420
0.315	1.1866	248	0.5710	39.1186
0.2988	1.4833	310	0.5689	45.5393
0.2893	1.7799	372	0.5588	37.1330
0.2607	2.0766	434	0.5569	36.0230
0.2532	2.3732	496	0.5575	35.7046
0.2599	2.6699	558	0.5543	34.4776
0.2716	2.9665	620	0.5545	35.1051

Framework versions

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