w2v2-base-pretrained_lr1e-4_at0.6_da1

This model is a fine-tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 4.4395
• 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: 0.0001
• train_batch_size: 32
• 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: 1000
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
23.3635	4.24	250	17.7956	1.0
7.0327	8.47	500	4.6430	1.0
3.2483	12.71	750	4.3311	1.0
3.196	16.95	1000	4.3287	1.0
3.1871	21.19	1250	4.1426	1.0
3.1775	25.42	1500	4.2391	1.0
3.1721	29.66	1750	4.2448	1.0
3.1701	33.9	2000	4.3298	1.0
3.1714	38.14	2250	4.3414	1.0
3.1674	42.37	2500	4.4173	1.0
3.1667	46.61	2750	4.3994	1.0
3.166	50.85	3000	4.4289	1.0
3.1657	55.08	3250	4.4452	1.0
3.1611	59.32	3500	4.4395	1.0

Framework versions

• Transformers 4.35.0
• Pytorch 2.0.0
• Datasets 2.14.6
• Tokenizers 0.14.1