hubert-base-libri-demo-feature_extractor_not_frozen_v3_25epochs_check

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: 0.1231
• Wer: 0.1112

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
4.3342	1.12	500	3.4935	1.0000
2.8802	2.24	1000	3.5637	1.0000
2.1866	3.36	1500	0.7219	0.6232
0.6141	4.48	2000	0.2954	0.3238
0.3328	5.61	2500	0.1810	0.2212
0.2251	6.73	3000	0.1377	0.1640
0.1861	7.85	3500	0.1270	0.1473
0.1671	8.97	4000	0.1173	0.1372
0.1496	10.09	4500	0.1218	0.1322
0.117	11.21	5000	0.1180	0.1268
0.1182	12.33	5500	0.1255	0.1257
0.0948	13.45	6000	0.1215	0.1221
0.0935	14.57	6500	0.1233	0.1217
0.0873	15.7	7000	0.1124	0.1209
0.0798	16.82	7500	0.1172	0.1185
0.0752	17.94	8000	0.1197	0.1171
0.0747	19.06	8500	0.1252	0.1171
0.0775	20.18	9000	0.1209	0.1149
0.0665	21.3	9500	0.1180	0.1133
0.0657	22.42	10000	0.1240	0.1122
0.0606	23.54	10500	0.1222	0.1110
0.0581	24.66	11000	0.1231	0.1112

Framework versions

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