xls-r-1b-bemgen-combined-model

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

• Loss: 0.2509
• Wer: 0.3923

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: 3e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.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: 500
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.1784	100	3.4413	1.0003
No log	0.3568	200	2.9149	1.0
No log	0.5352	300	0.7768	0.9235
No log	0.7136	400	0.6057	0.9047
5.3372	0.8921	500	0.4317	0.6720
5.3372	1.0696	600	0.3997	0.6704
5.3372	1.2480	700	0.3611	0.6405
5.3372	1.4264	800	0.3441	0.5603
5.3372	1.6048	900	0.2945	0.4914
0.6459	1.7832	1000	0.3041	0.4924
0.6459	1.9616	1100	0.2805	0.4681
0.6459	2.1392	1200	0.2774	0.5108
0.6459	2.3176	1300	0.2683	0.4254
0.6459	2.4960	1400	0.2644	0.4382
0.4599	2.6744	1500	0.2446	0.4142
0.4599	2.8528	1600	0.2473	0.4118
0.4599	3.0303	1700	0.2492	0.3961
0.4599	3.2087	1800	0.2467	0.4070
0.4599	3.3872	1900	0.2509	0.3923

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0