w2v2-base-pretrained_lr1e-4_at0.8_da0.8

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: 3.0247
• 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
• training_steps: 3500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
14.2507	6.76	250	3.8646	1.0
3.1822	13.51	500	3.1594	1.0
3.104	20.27	750	3.1306	1.0
3.0856	27.03	1000	3.1454	1.0
3.0745	33.78	1250	3.1263	1.0
3.0518	40.54	1500	3.0992	1.0
3.0403	47.3	1750	3.0938	1.0
3.023	54.05	2000	3.1334	1.0
3.0097	60.81	2250	3.0640	1.0
2.9931	67.57	2500	3.0890	1.0
2.9778	74.32	2750	3.0379	1.0
2.9657	81.08	3000	3.0294	1.0
2.9529	87.84	3250	3.0291	1.0
2.9447	94.59	3500	3.0247	1.0

Framework versions

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