whisper-new-nnat-10h-random-seed2025

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: 0.5099
• Wer: 23.3784

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: 22
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9403	0.2933	22	1.2296	44.5511
0.6265	0.5867	44	0.6971	28.6330
0.467	0.88	66	0.5845	24.6897
0.4243	1.1733	88	0.5499	23.7999
0.4077	1.4667	110	0.5357	23.1349
0.4168	1.76	132	0.5267	22.7977
0.4053	2.0533	154	0.5174	24.3525
0.3697	2.3467	176	0.5140	22.6010
0.3849	2.64	198	0.5113	23.2286
0.3839	2.9333	220	0.5099	23.3784

Framework versions

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