thelikhit/HuBERT-draft-version

hubert-test-model

This model is a fine-tuned version of facebook/hubert-base-ls960 on the gtzan dataset. It achieves the following results on the evaluation set:

• Loss: 1.5276
• Accuracy: 0.785

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: 4
• eval_batch_size: 4
• 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
No log	1.0	150	2.0494	0.29
No log	2.0	300	2.1993	0.19
No log	3.0	450	1.8439	0.44
1.9218	4.0	600	1.5277	0.48
1.9218	5.0	750	1.4164	0.475
1.9218	6.0	900	1.3641	0.63
1.2685	7.0	1050	1.1557	0.675
1.2685	8.0	1200	1.0935	0.72
1.2685	9.0	1350	1.0594	0.71
0.7151	10.0	1500	1.0119	0.735
0.7151	11.0	1650	1.0868	0.77
0.7151	12.0	1800	1.3736	0.75
0.7151	13.0	1950	1.2705	0.77
0.4135	14.0	2100	1.4052	0.76
0.4135	15.0	2250	1.3864	0.77
0.4135	16.0	2400	1.4296	0.785
0.2311	17.0	2550	1.5663	0.77
0.2311	18.0	2700	1.5310	0.78
0.2311	19.0	2850	1.4884	0.795
0.1408	20.0	3000	1.5276	0.785

Framework versions

• Transformers 4.46.2
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3