SLPL/Hubert-base-ShEMO

results

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

• Loss: 0.6161
• F1: 0.8335

Labels description

• 0 : anger
• 1 : happiness
• 2 : neutral
• 3 : sadness

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: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	F1
1.1127	1.0	154	0.9244	0.3968
0.6982	2.0	308	0.5642	0.6435
0.6246	3.0	462	0.5049	0.6273
0.5097	4.0	616	0.4282	0.7246
0.4496	5.0	770	0.3280	0.8158
0.4476	6.0	924	0.4663	0.7978
0.2212	7.0	1078	0.3253	0.8641
0.1548	8.0	1232	0.9445	0.7420
0.3829	9.0	1386	0.7194	0.7880
0.0773	10.0	1540	0.5301	0.8657
0.2481	11.0	1694	0.5321	0.8812
0.0597	12.0	1848	0.6161	0.8335

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1