wav2vec2-large-sw-cv-100hr-v6

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: 0.6098
• Model Preparation Time: 0.0065
• Wer: 0.3875
• Cer: 0.1346

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: 0.001
• train_batch_size: 64
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.2
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Model Preparation Time	Wer
2.1201	1.0	1040	0.1472	0.6354	0.0059	0.5377
0.4194	2.0	2080	0.1104	0.4653	0.0059	0.4050
0.3266	3.0	3120	0.1087	0.3807	0.0059	0.3726
0.2997	4.0	4160	0.0944	0.3834	0.0059	0.3445
0.2984	5.0	5200	0.0979	0.4147	0.0059	0.3493
0.3072	6.0	6240	0.1020	0.4111	0.0059	0.3583
0.3265	7.0	7280	0.1147	0.4162	0.0059	0.3947
0.3477	8.0	8320	0.4503	0.0065	0.3979	0.1191
0.384	9.0	9360	0.5286	0.0065	0.4552	0.1442
0.4269	10.0	10400	0.5312	0.0065	0.4761	0.1471
1.5371	11.0	11440	4.5667	0.0065	0.9999	0.9827
4.5758	12.0	12480	4.5229	0.0065	0.9999	0.9827
4.5739	13.0	13520	4.5302	0.0065	0.9999	0.9827
4.5753	14.0	14560	4.5329	0.0065	0.9999	0.9827
4.5736	15.0	15600	4.5338	0.0065	0.9999	0.9827
4.5724	16.0	16640	4.5463	0.0065	0.9999	0.9827
4.5738	17.0	17680	4.5220	0.0065	0.9999	0.9827

Framework versions

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