whisper-child-25

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: 0.4041
• Wer: 20.6394

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: 61
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.1164	0.1225	25	1.2201	38.0716
1.0913	0.2451	50	1.1869	37.5885
1.0308	0.3676	75	1.1215	36.4646
0.912	0.4902	100	1.0354	35.3105
0.8776	0.6127	125	0.9431	34.6496
0.7639	0.7353	150	0.8479	32.7541
0.7541	0.8578	175	0.7468	33.1332
0.6172	0.9804	200	0.6488	30.8485
0.6094	1.1029	225	0.5804	27.7217
0.5589	1.2255	250	0.5284	25.5578
0.5294	1.3480	275	0.4859	25.1552
0.5191	1.4706	300	0.4579	23.9172
0.5059	1.5931	325	0.4427	22.6021
0.4534	1.7157	350	0.4323	21.9512
0.4812	1.8382	375	0.4247	21.0924
0.4884	1.9608	400	0.4199	22.2565
0.4994	2.0833	425	0.4158	22.6188
0.4257	2.2059	450	0.4127	22.5652
0.4264	2.3284	475	0.4098	22.0452
0.4577	2.4510	500	0.4076	20.7065
0.4442	2.5735	525	0.4062	20.6629
0.4432	2.6961	550	0.4051	20.6394
0.4703	2.8186	575	0.4044	20.6428
0.4313	2.9412	600	0.4041	20.6394

Framework versions

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