whisper-child-100

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: 0.3725
• Wer: 18.3581

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: 73
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0203	0.1029	25	1.2205	38.0716
1.0014	0.2058	50	1.1907	37.7663
0.9488	0.3086	75	1.1275	36.7967
0.9003	0.4115	100	1.0389	35.3440
0.8202	0.5144	125	0.9320	34.5489
0.7525	0.6173	150	0.8142	32.4991
0.6621	0.7202	175	0.6837	32.2944
0.5552	0.8230	200	0.5817	28.6007
0.5226	0.9259	225	0.5223	25.6685
0.5076	1.0288	250	0.4793	24.2829
0.4432	1.1317	275	0.4476	22.3605
0.4251	1.2346	300	0.4288	20.9917
0.4478	1.3374	325	0.4167	20.4214
0.3996	1.4403	350	0.4083	20.2939
0.4385	1.5432	375	0.4022	20.0020
0.4277	1.6461	400	0.3969	19.7403
0.4122	1.7490	425	0.3925	19.6464
0.4094	1.8519	450	0.3886	19.1096
0.4097	1.9547	475	0.3855	19.2740
0.3878	2.0576	500	0.3827	18.5594
0.4217	2.1605	525	0.3805	19.0157
0.3783	2.2634	550	0.3786	18.9251
0.4282	2.3663	575	0.3769	18.9016
0.4174	2.4691	600	0.3756	19.2002
0.3832	2.5720	625	0.3745	19.1733
0.3477	2.6749	650	0.3736	18.6902
0.3844	2.7778	675	0.3730	18.3581
0.4023	2.8807	700	0.3727	18.5695
0.3951	2.9835	725	0.3725	18.3581

Framework versions

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