wav2vec2_xls_r_300m_FL_xh_5hr_v2

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the fleurs dataset. It achieves the following results on the evaluation set:

• Loss: 0.7307
• Wer: 0.6436
• Cer: 0.1450

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• 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: 500
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
10.798	1.1905	100	4.1013	1.0	1.0
3.5497	2.3810	200	3.1255	1.0	1.0
3.0281	3.5714	300	2.9738	1.0	1.0
2.8915	4.7619	400	2.6340	1.0	0.9511
1.6204	5.9524	500	0.9010	0.8642	0.2181
0.7243	7.1429	600	0.7041	0.7730	0.1947
0.4925	8.3333	700	0.6179	0.7049	0.1612
0.3664	9.5238	800	0.6255	0.6972	0.1568
0.281	10.7143	900	0.6373	0.6767	0.1515
0.2211	11.9048	1000	0.6508	0.6683	0.1501
0.1792	13.0952	1100	0.6884	0.6662	0.1510
0.1517	14.2857	1200	0.6849	0.6485	0.1448
0.1356	15.4762	1300	0.7114	0.6594	0.1488
0.1222	16.6667	1400	0.7200	0.6506	0.1463
0.1106	17.8571	1500	0.7253	0.6478	0.1450
0.1065	19.0476	1600	0.7223	0.6450	0.1454

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1