wav2vec2_xls_r_300m_BIG-C_Bemba_50hr_v3

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

• Loss: 0.4961
• Model Preparation Time: 0.006
• Wer: 0.4359
• Cer: 0.1111

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: 32
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Cer
3.6745	1.0	394	inf	0.006	0.6294	0.1564
0.6444	2.0	788	inf	0.006	0.4910	0.1283
0.5606	3.0	1182	inf	0.006	0.4686	0.1203
0.521	4.0	1576	inf	0.006	0.4644	0.1227
0.4858	5.0	1970	inf	0.006	0.4451	0.1157
0.4571	6.0	2364	inf	0.006	0.4622	0.1212
0.4348	7.0	2758	inf	0.006	0.4716	0.1417
0.7234	8.0	3152	inf	0.006	0.9825	0.3581
1.7422	9.0	3546	inf	0.006	0.9813	0.3970
1.819	10.0	3940	inf	0.006	0.9996	0.5327
1.0484	11.0	4334	inf	0.006	0.5059	0.1368
1.9919	12.0	4728	nan	0.006	0.9999	0.6240
0.8944	13.0	5122	nan	0.006	0.9999	0.6240
2.7123	14.0	5516	nan	0.006	1.0	1.0
0.0	15.0	5910	nan	0.006	1.0	1.0
0.0	16.0	6304	nan	0.006	1.0	1.0
0.0	17.0	6698	nan	0.006	1.0	1.0
0.0	18.0	7092	nan	0.006	1.0	1.0
0.0	19.0	7486	nan	0.006	1.0	1.0
0.0	20.0	7880	nan	0.006	1.0	1.0

Framework versions

• Transformers 4.44.0
• Pytorch 2.2.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1