whisper-child-50

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: 0.3919
• Wer: 20.2872

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: 65
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0614	0.1157	25	1.2199	38.0783
1.0412	0.2315	50	1.1881	37.6489
0.9789	0.3472	75	1.1234	36.6021
0.9321	0.4630	100	1.0316	35.1931
0.8436	0.5787	125	0.9318	34.7133
0.7731	0.6944	150	0.8271	32.6568
0.706	0.8102	175	0.7151	31.6100
0.6152	0.9259	200	0.6261	29.7816
0.6491	1.0417	225	0.5572	26.0845
0.506	1.1574	250	0.5066	24.5546
0.5038	1.2731	275	0.4675	23.0315
0.4662	1.3889	300	0.4427	22.6524
0.4616	1.5046	325	0.4296	22.3471
0.476	1.6204	350	0.4208	22.1861
0.456	1.7361	375	0.4139	21.3306
0.4899	1.8519	400	0.4091	21.2400
0.4576	1.9676	425	0.4053	20.6965
0.4289	2.0833	450	0.4021	20.5321
0.4594	2.1991	475	0.3995	20.3878
0.4288	2.3148	500	0.3971	20.0087
0.4221	2.4306	525	0.3954	20.3643
0.4239	2.5463	550	0.3941	20.3711
0.4496	2.6620	575	0.3931	20.3476
0.4303	2.7778	600	0.3923	20.3073
0.458	2.8935	625	0.3919	20.2872

Framework versions

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