wav2vec2-large-sw-cv-100hr-v3

This model is a fine-tuned version of facebook/wav2vec2-large on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6016
• Model Preparation Time: 0.0041
• Wer: 0.4019
• Cer: 0.1436

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: 0.0005
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.15
• num_epochs: 120
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Cer
1.6262	0.9998	2079	0.5262	0.0041	0.5289	0.1392
0.3281	2.0	4159	0.4055	0.0041	0.4037	0.1134
0.265	2.9998	6238	0.3537	0.0041	0.3599	0.0974
0.2592	4.0	8318	0.3882	0.0041	0.3790	0.1157
0.27	4.9998	10397	0.4337	0.0041	0.3919	0.1124
0.3063	6.0	12477	0.4226	0.0041	0.4094	0.1204
2.7704	6.9998	14556	2.8607	0.0041	1.0	1.0
2.86	8.0	16636	2.8618	0.0041	1.0	1.0
2.861	8.9998	18715	2.8596	0.0041	1.0	1.0
2.8597	10.0	20795	2.8618	0.0041	1.0	1.0
2.8611	10.9998	22874	2.8581	0.0041	1.0	1.0
2.8597	12.0	24954	2.8571	0.0041	1.0	1.0
2.861	12.9998	27033	2.8568	0.0041	1.0	1.0

Framework versions

• Transformers 4.44.0
• Pytorch 2.1.0+cu118
• Datasets 2.20.0
• Tokenizers 0.19.1