w2v2-base-pretrained_lr5e-5_at0.8_da0.025

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: 3.9947
• Wer: 1.0205

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: 5e-05
• 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
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
36.9594	125.0	250	12.4193	1.0
5.8739	250.0	500	3.2963	1.0
3.1532	375.0	750	3.1231	1.0
3.0639	500.0	1000	3.1060	1.0
3.013	625.0	1250	3.1074	1.0
2.9755	750.0	1500	3.1334	1.0
2.94	875.0	1750	3.1535	1.0
2.8802	1000.0	2000	3.0883	1.0
2.474	1125.0	2250	2.8111	1.0009
1.3378	1250.0	2500	3.0612	1.0038
0.7759	1375.0	2750	3.4681	1.0068
0.594	1500.0	3000	3.6791	1.0337
0.5034	1625.0	3250	3.8161	1.0226
0.4518	1750.0	3500	3.8285	1.0081
0.4139	1875.0	3750	3.9486	1.0201
0.3953	2000.0	4000	3.9947	1.0205

Framework versions

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