distilhubert-gtzan-loraAL-dropout0.5-split3

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

• Loss: 1.7145
• Accuracy: 0.84

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: 0.001
• train_batch_size: 16
• eval_batch_size: 16
• 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: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
2.2071	1.0	169	1.4833	0.46
1.1886	2.0	338	1.1455	0.6367
0.8256	3.0	507	0.9532	0.7367
0.6972	4.0	676	1.0910	0.7
0.6284	5.0	845	0.7128	0.7667
0.5584	6.0	1014	0.7270	0.8
0.4614	7.0	1183	0.7947	0.8067
0.3895	8.0	1352	0.8538	0.7967
0.3605	9.0	1521	1.1667	0.7633
0.2569	10.0	1690	1.0894	0.8133
0.2078	11.0	1859	1.2071	0.8033
0.159	12.0	2028	1.4872	0.78
0.1177	13.0	2197	1.2216	0.8467
0.0692	14.0	2366	1.4827	0.82
0.0624	15.0	2535	1.5108	0.8433
0.0406	16.0	2704	1.6933	0.8333
0.0136	17.0	2873	1.6956	0.8333
0.0148	18.0	3042	1.7178	0.8367
0.0065	19.0	3211	1.7237	0.8333
0.0078	20.0	3380	1.7145	0.84

Framework versions

• PEFT 0.15.2
• Transformers 4.51.3
• Pytorch 2.6.0
• Datasets 3.3.2
• Tokenizers 0.21.0