mitro99/distilhubert-finetuned-gtzan_batch8_grad4_cosine

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:

• Accuracy: 0.88
• Loss: 0.4331

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
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Accuracy	Validation Loss
2.2693	0.99	28	0.31	2.2480
1.9782	1.98	56	0.45	1.8990
1.6438	2.97	84	0.62	1.5180
1.3307	4.0	113	0.73	1.2206
1.133	4.99	141	0.76	0.9961
0.9384	5.98	169	0.78	0.8889
0.8668	6.97	197	0.79	0.7543
0.674	8.0	226	0.79	0.7433
0.5997	8.99	254	0.83	0.6194
0.5195	9.98	282	0.91	0.5685
0.401	10.97	310	0.91	0.5144
0.3151	12.0	339	0.87	0.4775
0.2653	12.99	367	0.88	0.4984
0.2182	13.98	395	0.88	0.4337
0.2036	14.97	423	0.89	0.4657
0.1925	16.0	452	0.89	0.4222
0.1807	16.99	480	0.87	0.4512
0.1626	17.98	508	0.88	0.4247
0.1388	18.97	536	0.88	0.4324
0.1718	19.82	560	0.88	0.4331

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.1+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0