w2v2-libri-10min

This model is a fine-tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.9408
• Wer: 0.5837

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 2500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.8824	62.5	250	2.9255	1.0
0.7061	125.0	500	1.5759	0.6321
0.066	187.5	750	1.7474	0.6183
0.0326	250.0	1000	1.7446	0.6224
0.0205	312.5	1250	1.8737	0.6252
0.0445	375.0	1500	1.9835	0.6210
0.0084	437.5	1750	1.8829	0.6141
0.0068	500.0	2000	1.9136	0.6058
0.0037	562.5	2250	1.8990	0.5864
0.003	625.0	2500	1.9408	0.5837

Framework versions

• Transformers 4.41.0
• Pytorch 2.3.0+cu121
• Datasets 1.18.1
• Tokenizers 0.19.1