wav2vec2-1b-Elderly6

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: 0.3776
• Cer: 10.3853

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: 1
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 16
• 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
9.195	0.2580	200	3.1758	50.8870
1.6332	0.5161	400	1.4825	34.3280
1.1542	0.7741	600	1.0327	24.9765
0.9689	1.0322	800	0.9404	22.8090
0.8334	1.2902	1000	0.7965	19.8250
0.7615	1.5483	1200	0.7948	19.2845
0.7499	1.8063	1400	0.7138	18.9203
0.6314	2.0643	1600	0.7851	19.9659
0.5657	2.3224	1800	0.6697	17.9335
0.5533	2.5804	2000	0.7101	18.7441
0.5326	2.8385	2200	0.6084	16.3240
0.4909	3.0965	2400	0.5517	15.5898
0.4061	3.3546	2600	0.5080	13.9803
0.3904	3.6126	2800	0.4536	12.5352
0.3628	3.8707	3000	0.4773	12.7232
0.3109	4.1287	3200	0.4460	12.0242
0.2835	4.3867	3400	0.4071	11.3311
0.2507	4.6448	3600	0.3907	10.6497
0.248	4.9028	3800	0.3776	10.3853

Framework versions

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