whisper-new-nnat-20h-meta-seed333

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: 0.4644
• Wer: 19.9972

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: 39
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.2682	0.3	39	0.9069	45.0616
0.6645	0.6	78	0.5525	24.3245
0.5616	0.9	117	0.5117	22.0344
0.5156	1.2	156	0.4948	21.1867
0.4889	1.5	195	0.4835	21.0228
0.4997	1.8	234	0.4739	20.5732
0.4574	2.1	273	0.4687	20.3718
0.4434	2.4	312	0.4669	20.6013
0.4495	2.7	351	0.4651	20.7886
0.4197	3.0	390	0.4644	19.9972

Framework versions

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