distilhubert-debiasing-age

This model is a fine-tuned version of ntu-spml/distilhubert on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 1.8820
• Accuracy: 0.6702

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use 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_ratio: 0.1
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
1.2915	1.0	117	1.4201	0.6064
1.3727	2.0	234	1.7079	0.6064
1.1821	3.0	351	1.6523	0.5851
1.3401	4.0	468	1.4616	0.6383
1.6345	5.0	585	1.5092	0.6170
1.1763	6.0	702	1.6093	0.6277
0.999	7.0	819	1.4447	0.6277
0.9842	8.0	936	1.5173	0.6702
0.9366	9.0	1053	1.8773	0.6809
0.9529	10.0	1170	1.8331	0.6489
1.2192	11.0	1287	2.0470	0.6702
0.8482	12.0	1404	1.9989	0.6809
0.9902	13.0	1521	2.3879	0.6383
1.0078	14.0	1638	2.1982	0.6809
0.9427	15.0	1755	1.9457	0.6596
0.9801	16.0	1872	1.9722	0.6702
0.9372	17.0	1989	1.9988	0.6596
0.9671	18.0	2106	1.8085	0.7128
0.9031	19.0	2223	1.8938	0.6702
0.8846	20.0	2340	1.8820	0.6702

Framework versions

• Transformers 4.50.3
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.0