wav2vec2_xls_r_300m_BIG-C_Bemba_20hr_v5

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: 0.5637
• Model Preparation Time: 0.0067
• Wer: 0.4975
• Cer: 0.1239

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
6.9607	1.0	154	inf	0.0067	1.0	1.0
2.4999	2.0	308	inf	0.0067	0.9981	0.2571
0.8598	3.0	462	inf	0.0067	0.6688	0.1653
0.6842	4.0	616	inf	0.0067	0.7770	0.1976
0.6052	5.0	770	inf	0.0067	0.5482	0.1443
0.557	6.0	924	inf	0.0067	0.5141	0.1334
0.5169	7.0	1078	inf	0.0067	0.6443	0.1941
0.4708	8.0	1232	inf	0.0067	0.5685	0.1573
0.4684	9.0	1386	inf	0.0067	0.5414	0.1489
0.5716	10.0	1540	inf	0.0067	0.6515	0.1929
1.1996	11.0	1694	inf	0.0067	0.9517	0.3363
1.4771	12.0	1848	inf	0.0067	0.9655	0.3621
1.3761	13.0	2002	inf	0.0067	0.8545	0.2684
1.6064	14.0	2156	inf	0.0067	1.0	0.3888
1.8647	15.0	2310	inf	0.0067	0.9999	0.5035
1.7455	16.0	2464	inf	0.0067	1.0	0.3958
1.6089	17.0	2618	inf	0.0067	1.0	0.3634
1.4697	18.0	2772	inf	0.0067	0.9999	0.3466
1.4016	19.0	2926	inf	0.0067	1.0	0.3309
1.3563	20.0	3080	inf	0.0067	1.0	0.3232
1.3217	21.0	3234	inf	0.0067	0.9999	0.3116

Framework versions

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