hubert-base-libri-demo-feature_extractor_frozen

This model is a fine-tuned version of facebook/hubert-base-ls960 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1217
• Wer: 0.3031

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: 0.00015
• train_batch_size: 64
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 3000
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.446	1.12	500	3.5285	0.9999
2.8792	2.24	1000	3.2540	0.9999
2.8504	3.36	1500	3.0608	0.9999
1.2084	4.48	2000	0.3020	0.6763
0.3281	5.61	2500	0.1665	0.4887
0.2158	6.73	3000	0.1389	0.4071
0.179	7.85	3500	0.1294	0.3710
0.163	8.97	4000	0.1199	0.3592
0.1461	10.09	4500	0.1177	0.3485
0.1138	11.21	5000	0.1120	0.3374
0.1154	12.33	5500	0.1215	0.3307
0.0919	13.45	6000	0.1182	0.3279
0.0911	14.57	6500	0.1194	0.3286
0.0856	15.7	7000	0.1163	0.3185
0.0786	16.82	7500	0.1154	0.3193
0.0738	17.94	8000	0.1124	0.3122
0.0738	19.06	8500	0.1185	0.3105
0.0767	20.18	9000	0.1208	0.3061
0.0664	21.3	9500	0.1211	0.3050
0.0654	22.42	10000	0.1189	0.3039
0.0606	23.54	10500	0.1235	0.3041
0.0584	24.66	11000	0.1217	0.3031

Framework versions

• Transformers 4.30.0.dev0
• Pytorch 2.0.1
• Datasets 2.12.1.dev0
• Tokenizers 0.13.3