ai-light-dance_singing2_ft_wav2vec2-large-xlsr-53

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the /WORKSPACE/ASANTE/AI-LIGHT-DANCE_DATASETS/AI_LIGHT_DANCE.PY - ONSET-SINGING2 dataset. It achieves the following results on the evaluation set:

• Loss: 1.7583
• Wer: 0.9386

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: 4e-06
• train_batch_size: 10
• eval_batch_size: 10
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 160
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
27.4755	1.0	112	23.2618	1.0
5.5145	2.0	224	5.2213	1.0
4.2211	3.0	336	4.1673	1.0
3.8386	4.0	448	3.8253	1.0
3.5531	5.0	560	3.6286	1.0
3.5215	6.0	672	3.4762	0.9864
3.3493	7.0	784	3.3549	0.9847
3.1264	8.0	896	3.1797	0.9759
2.7557	9.0	1008	2.8703	0.9865
2.6345	10.0	1120	2.6736	0.9970
2.4297	11.0	1232	2.5638	1.0337
2.3057	12.0	1344	2.3680	0.9839
2.1436	13.0	1456	2.2367	0.9648
2.0856	14.0	1568	2.1635	0.9586
2.0035	15.0	1680	2.0945	0.9645
1.9134	16.0	1792	2.0395	0.9630
1.9443	17.0	1904	2.0017	0.9401
1.8988	18.0	2016	1.9514	0.9493
1.8141	19.0	2128	1.9111	0.9475
1.8344	20.0	2240	1.8790	0.9395
1.7775	21.0	2352	1.8616	0.9503
1.7517	22.0	2464	1.8333	0.9433
1.7037	23.0	2576	1.8156	0.9372
1.7158	24.0	2688	1.7961	0.9482
1.7111	25.0	2800	1.7817	0.9422
1.69	26.0	2912	1.7819	0.9430
1.6889	27.0	3024	1.7721	0.9386
1.6546	28.0	3136	1.7647	0.9453
1.6542	29.0	3248	1.7653	0.9375
1.647	30.0	3360	1.7583	0.9386

Framework versions

• Transformers 4.21.0.dev0
• Pytorch 1.9.1+cu102
• Datasets 2.3.3.dev0
• Tokenizers 0.12.1