XLS-R_Jibbali_lang

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: 0.1752
• Wer: 0.1926
• Cer: 0.0770

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
16.5992	0.99	56	15.4050	1.0	0.9812
3.79	2.0	113	3.3988	1.0	0.9812
3.1872	2.99	169	3.1498	1.0	0.9812
3.1705	4.0	226	3.1354	1.0	0.9812
3.1147	4.99	282	3.0947	1.0	0.9812
3.0616	6.0	339	2.9447	1.0	0.9460
2.8239	6.99	395	2.6661	1.0	0.9106
1.5494	8.0	452	1.0992	0.8684	0.3804
0.5291	8.99	508	0.2822	0.3026	0.1004
0.2022	10.0	565	0.2019	0.2080	0.0665
0.1721	10.99	621	0.2067	0.2032	0.0841
0.1705	12.0	678	0.1968	0.1996	0.0728
0.0989	12.99	734	0.2038	0.1955	0.0821
0.1299	14.0	791	0.1814	0.1963	0.0837
0.1352	14.99	847	0.1896	0.1941	0.0768
0.0487	16.0	904	0.1951	0.1933	0.0749
0.1412	16.99	960	0.1650	0.1970	0.0818
0.1027	18.0	1017	0.1720	0.1941	0.0783
0.0791	18.99	1073	0.1730	0.1933	0.0767
0.0406	19.82	1120	0.1752	0.1926	0.0770

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.17.0
• Tokenizers 0.15.2