wav2vec2_xls_r_300m_BIG-C_Bemba_1hr_v2

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: 14.9650
• Model Preparation Time: 0.0065
• Wer: 1.0
• Cer: 1.0883

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: 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: 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
15.5971	0.9677	15	14.9495	0.0065	1.0	1.0688
14.3431	2.0	31	14.4988	0.0065	1.0	0.9944
14.6096	2.9677	46	13.5477	0.0065	1.0	0.9877
12.0537	4.0	62	10.5808	0.0065	1.0	1.0
8.4919	4.9677	77	6.4485	0.0065	1.0	1.0
5.2708	6.0	93	4.7793	0.0065	1.0	1.0
4.5315	6.9677	108	4.1769	0.0065	1.0	1.0
3.8136	8.0	124	3.8360	0.0065	1.0	1.0
3.813	8.9677	139	3.6393	0.0065	1.0	1.0
3.4064	10.0	155	3.4606	0.0065	1.0	1.0
3.4865	10.9677	170	3.3233	0.0065	1.0	1.0

Framework versions

• Transformers 4.43.4
• Pytorch 2.2.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1