distilhubert-finetuned-gtzan

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: 0.80
• Accuracy: 0.89
• Macro F1: 0.8957

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: 3e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• 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: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Macro F1
2.2641	1.0	450	2.2228	0.45	0.3545
1.8857	2.0	900	1.8527	0.56	0.4640
1.6008	3.0	1350	1.4108	0.68	0.6543
1.1742	4.0	1800	1.1191	0.72	0.6910
0.9257	5.0	2250	0.8738	0.81	0.8026
0.7766	6.0	2700	0.6843	0.82	0.8216
0.4557	7.0	3150	0.6263	0.82	0.8170
0.3278	8.0	3600	0.5555	0.84	0.8285
0.3262	9.0	4050	0.5471	0.87	0.8641
0.2547	10.0	4500	0.7685	0.81	0.8043
0.2713	11.0	4950	0.6107	0.89	0.8782
0.1555	12.0	5400	0.6796	0.88	0.8767
0.1983	13.0	5850	0.8251	0.86	0.8466
0.1146	14.0	6300	0.8332	0.85	0.8341
0.1164	15.0	6750	0.8060	0.89	0.8957
0.0103	16.0	7200	1.1965	0.84	0.8345
0.0247	17.0	7650	1.1232	0.86	0.8539
0.0037	18.0	8100	1.0172	0.87	0.8665
0.0021	19.0	8550	0.9868	0.88	0.8734
0.0594	20.0	9000	1.4043	0.86	0.8591

Framework versions

• Transformers 4.53.3
• Pytorch 2.9.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.1