wav2vec2-base-checkpoint-5

This model is a fine-tuned version of jiobiala24/wav2vec2-base-checkpoint-4 on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.9849
• Wer: 0.3354

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: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3947	1.96	1000	0.5749	0.3597
0.2856	3.93	2000	0.6212	0.3479
0.221	5.89	3000	0.6280	0.3502
0.1755	7.86	4000	0.6517	0.3526
0.1452	9.82	5000	0.7115	0.3481
0.1256	11.79	6000	0.7687	0.3509
0.1117	13.75	7000	0.7785	0.3490
0.0983	15.72	8000	0.8115	0.3442
0.0877	17.68	9000	0.8290	0.3429
0.0799	19.65	10000	0.8517	0.3412
0.0733	21.61	11000	0.9370	0.3448
0.066	23.58	12000	0.9157	0.3410
0.0623	25.54	13000	0.9673	0.3377
0.0583	27.5	14000	0.9804	0.3348
0.0544	29.47	15000	0.9849	0.3354

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu111
• Datasets 1.13.3
• Tokenizers 0.10.3