w2v2-base-pretrained_lr1e-4_at0.7_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: 6.2022
• Wer: 1.0580

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
18.007	4.46	250	4.3814	1.0
3.2628	8.93	500	3.8568	1.0
3.1079	13.39	750	3.9328	1.0
1.4764	17.86	1000	2.7696	1.0384
0.2321	22.32	1250	4.2808	1.0507
0.1312	26.79	1500	4.8707	1.0529
0.0793	31.25	1750	5.2587	1.0558
0.0546	35.71	2000	5.6739	1.0541
0.0401	40.18	2250	5.7379	1.0494
0.0303	44.64	2500	5.8382	1.0558
0.0255	49.11	2750	6.0859	1.0567
0.0223	53.57	3000	6.0789	1.0558
0.019	58.04	3250	6.2022	1.0580

Framework versions

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