galsenai/hubert-base-ls960-ft-waxal-keyword-spotting

hubert-base-ls960

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

• Loss: 2.1857
• Accuracy: 0.6442
• Precision: 0.8369
• F1: 0.7121

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: 30
• eval_batch_size: 30
• seed: 0
• gradient_accumulation_steps: 4
• total_train_batch_size: 120
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 32.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	F1
4.523	2.53	500	5.1547	0.0205	0.0047	0.0037
3.4187	5.05	1000	4.6287	0.0337	0.0256	0.0163
2.3533	7.58	1500	4.2550	0.0944	0.1033	0.0641
1.7145	10.1	2000	3.9540	0.1095	0.2091	0.0964
1.3245	12.63	2500	3.8557	0.1758	0.3609	0.1859
1.0729	15.15	3000	3.7411	0.2247	0.4918	0.2537
0.8955	17.68	3500	3.2683	0.3789	0.6162	0.4256
0.7697	20.2	4000	2.8749	0.4612	0.7106	0.5171
0.6864	22.73	4500	2.7251	0.5169	0.7437	0.5779
0.6061	25.25	5000	2.5061	0.5631	0.8043	0.6335
0.5777	27.78	5500	2.2830	0.6177	0.8183	0.6837
0.5304	30.3	6000	2.1857	0.6442	0.8369	0.7121

Framework versions

• Transformers 4.27.0.dev0
• Pytorch 1.11.0+cu113
• Datasets 2.9.1.dev0
• Tokenizers 0.13.2