exp2-whisper-child-classified

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: 0.4059
• Wer: 19.8242

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: 1e-05
• train_batch_size: 48
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 54
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0368	0.1397	25	1.2168	38.0313
0.9962	0.2793	50	1.1738	37.1423
1.0082	0.4190	75	1.0938	36.2465
0.8819	0.5587	100	0.9987	34.8375
0.8438	0.6983	125	0.8940	34.3174
0.7026	0.8380	150	0.7810	32.3984
0.6271	0.9777	175	0.6621	32.2709
0.5367	1.1173	200	0.5746	29.4729
0.5252	1.2570	225	0.5207	26.1683
0.4623	1.3966	250	0.4818	23.5750
0.4399	1.5363	275	0.4557	23.7394
0.4141	1.6760	300	0.4407	21.9814
0.4084	1.8156	325	0.4310	21.3742
0.416	1.9553	350	0.4240	20.8105
0.3979	2.0950	375	0.4190	20.6696
0.3963	2.2346	400	0.4152	20.0758
0.4103	2.3743	425	0.4119	19.5525
0.4087	2.5140	450	0.4094	19.9349
0.4359	2.6536	475	0.4076	19.8980
0.4066	2.7933	500	0.4065	19.8544
0.3824	2.9330	525	0.4059	19.8242

Framework versions

• PEFT 0.16.0
• Transformers 4.52.0
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.2