distilhubert-finetuned-accents

This model is a fine-tuned version of ntu-spml/distilhubert on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 2.0748
• Accuracy: 0.2708

Model description

• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.7
• num_epochs: 14
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
2.4778	1.0	48	2.4807	0.0938
2.4779	2.0	96	2.4651	0.1042
2.4751	3.0	144	2.4365	0.1042
2.3777	4.0	192	2.4187	0.1042
2.3786	5.0	240	2.4050	0.1458
2.3754	6.0	288	2.3446	0.1458
2.1556	7.0	336	2.2284	0.2083
2.1062	8.0	384	2.1533	0.2188
2.0081	9.0	432	2.0765	0.2292
1.813	10.0	480	2.0671	0.2083
1.74	11.0	528	1.9977	0.3021
1.4795	12.0	576	2.0588	0.2396
1.298	13.0	624	2.0652	0.3021
1.2578	14.0	672	2.0748	0.2708

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0