wav2vec2-base-cer

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: 0.0018
• Cer: 0.0718

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: 64
• 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: 150
• num_epochs: 200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
6.646	15.38	200	1.9010	0.6387
0.6207	30.77	400	0.0849	0.1757
0.0527	46.15	600	0.0643	0.1386
0.0325	61.54	800	0.0117	0.0888
0.0156	76.92	1000	0.0101	0.1148
0.0081	92.31	1200	0.0042	0.1255
0.0057	107.69	1400	0.0036	0.1284
0.0058	123.08	1600	0.0066	0.0891
0.0066	138.46	1800	0.0028	0.0926
0.0049	153.85	2000	0.0026	0.0391
0.0044	169.23	2200	0.0020	0.0574
0.0024	184.62	2400	0.0018	0.0745
0.0023	200.0	2600	0.0018	0.0718

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.2+cu121
• Datasets 2.14.5
• Tokenizers 0.15.2