wavlm-basic_s-f-c_8batch_5sec_0.0001lr_unfrozen

This model is a fine-tuned version of microsoft/wavlm-large on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8095
• Accuracy: 0.85
• F1: 0.8383

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: 0.0001
• 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: linear
• lr_scheduler_warmup_ratio: 0.003
• num_epochs: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
2.2489	0.99	47	2.3092	0.1	0.0182
1.8953	2.0	95	2.1986	0.2	0.0807
1.6269	2.99	142	2.0505	0.2667	0.1554
1.4844	4.0	190	1.7348	0.4333	0.3482
1.2047	4.99	237	1.3970	0.5833	0.4907
1.005	6.0	285	1.3947	0.6	0.4957
0.8541	6.99	332	1.0432	0.65	0.5830
0.7027	8.0	380	1.0033	0.7333	0.6992
0.72	8.99	427	0.9982	0.7833	0.7657
0.5461	10.0	475	1.1170	0.6833	0.6571
0.4415	10.99	522	0.9240	0.75	0.7402
0.4022	12.0	570	0.9522	0.7667	0.7488
0.3664	12.99	617	0.8290	0.8333	0.8253
0.3592	14.0	665	1.0270	0.75	0.7313
0.2985	14.99	712	1.0835	0.7667	0.7591
0.2565	16.0	760	0.9175	0.8167	0.8090
0.2887	16.99	807	0.8095	0.85	0.8383
0.3038	18.0	855	0.8871	0.7833	0.7763
0.242	18.99	902	0.8786	0.8	0.7875
0.1994	20.0	950	1.0309	0.7833	0.7656
0.1569	20.99	997	1.0706	0.8	0.7886
0.1637	22.0	1045	0.9650	0.8333	0.8249

Framework versions

• Transformers 4.28.1
• Pytorch 2.0.0+cu118
• Datasets 2.12.0
• Tokenizers 0.13.3