wav2vec-sq-3.0

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.9327
• Wer: 0.6953

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: 1e-05
• train_batch_size: 32
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• training_steps: 2500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.166	0.2657	250	2.9837	1.0
5.9696	0.5313	500	2.8618	1.0
5.1332	0.7970	750	2.0889	0.9683
3.6118	1.0627	1000	1.4413	0.8753
2.6843	1.3284	1250	1.1732	0.8004
2.3974	1.5940	1500	1.0998	0.7510
2.2139	1.8597	1750	1.0167	0.7115
2.0901	2.1254	2000	0.9646	0.7069
2.0002	2.3911	2250	0.9416	0.6906
1.9739	2.6567	2500	0.9327	0.6953

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0