xls-r-1b-bem-natbed-combined-model

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the NATBED - BEM dataset. It achieves the following results on the evaluation set:

• Loss: 0.7026
• Wer: 0.7512

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.0296	0.2503	100	2.8071	1.0
1.5792	0.5006	200	1.1184	0.9645
1.1422	0.7509	300	1.0389	0.9606
0.9883	1.0013	400	1.0494	0.9989
0.8999	1.2516	500	0.8692	0.8683
0.9135	1.5019	600	0.8564	0.8430
0.8898	1.7522	700	0.8451	0.8522
0.9089	2.0025	800	0.8857	0.8485
0.8292	2.2528	900	0.8662	0.8580
0.7921	2.5031	1000	0.7964	0.7969
0.7983	2.7534	1100	0.7896	0.7951
0.7946	3.0038	1200	0.7667	0.7947
0.7488	3.2541	1300	0.8180	0.8495
0.7428	3.5044	1400	0.7548	0.7688
0.7256	3.7547	1500	0.7258	0.7596
0.741	4.0050	1600	0.7665	0.7718
0.6775	4.2553	1700	0.7922	0.7775
0.6795	4.5056	1800	0.7026	0.7512
0.683	4.7559	1900	0.7051	0.7225
0.6838	5.0063	2000	0.7196	0.7503
0.6005	5.2566	2100	0.7032	0.7424

Framework versions

• Transformers 4.46.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0