wav2vec2-base-sep28k-1666samples-v2

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

• Loss: 1.7237
• Wer: 0.3930

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.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: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.4445	2.9940	500	2.9543	1.0
2.9087	5.9880	1000	2.5961	1.0
1.5399	8.9820	1500	1.2670	0.5875
0.6604	11.9760	2000	1.2157	0.4817
0.394	14.9701	2500	1.3348	0.4480
0.2602	17.9641	3000	1.4616	0.4252
0.1881	20.9581	3500	1.5525	0.4129
0.139	23.9521	4000	1.6597	0.4017
0.1099	26.9461	4500	1.7237	0.3930

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0