hubert-base-libri-demo-feature_extractor_not_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: 3.8505
• Wer: 1.0

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
4.3787	1.12	500	3.5514	1.0
2.8772	2.24	1000	3.6079	1.0
2.8665	3.36	1500	4.0086	1.0
2.864	4.48	2000	4.0925	1.0
2.8606	5.61	2500	3.8393	1.0
2.8589	6.73	3000	3.7722	1.0
2.8603	7.85	3500	3.8588	1.0
2.8685	8.97	4000	3.8065	1.0
2.8602	10.09	4500	3.8807	1.0
2.858	11.21	5000	3.8724	1.0
2.8582	12.33	5500	3.8091	1.0
2.8579	13.45	6000	3.6785	1.0
2.8569	14.57	6500	3.7339	1.0
2.8574	15.7	7000	3.7972	1.0
2.8564	16.82	7500	3.8758	1.0
2.8569	17.94	8000	3.9114	1.0
2.8569	19.06	8500	3.9208	1.0
2.8565	20.18	9000	3.9229	1.0
2.8571	21.3	9500	3.8876	1.0
2.8569	22.42	10000	3.8732	1.0
2.8557	23.54	10500	3.8587	1.0
2.8569	24.66	11000	3.8505	1.0

Framework versions

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