wav2vec2-1b-E50

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5653
• Cer: 14.0038

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: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
10.3076	0.5161	200	3.6837	66.2829
1.6559	1.0323	400	1.1986	25.7695
0.9077	1.5484	600	0.9908	22.8912
0.7297	2.0645	800	0.7952	19.4549
0.5331	2.5806	1000	0.7366	18.3506
0.4468	3.0968	1200	0.7078	17.0465
0.3334	3.6129	1400	0.6080	15.1727
0.2825	4.1290	1600	0.5928	14.4796
0.2049	4.6452	1800	0.5653	14.0038

Framework versions

• Transformers 4.45.2
• Pytorch 2.3.1.post100
• Datasets 2.19.1
• Tokenizers 0.20.1