wav2vec2-xls-r-1b-luxembourgish-38h-11k-steps

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

• Loss: 0.5388
• Wer: 53.3282
• Cer: 15.6751

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: 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: 3000
• training_steps: 11000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.627	1.4286	1000	0.3266	39.7841	11.5375
0.9399	2.8571	2000	0.5695	46.6975	13.7326
0.7264	4.2857	3000	0.4885	48.5736	14.8389
0.7351	5.7143	4000	0.5389	53.3539	15.6838
0.759	7.1429	5000	0.5388	53.4567	15.6620
0.7855	8.5714	6000	0.5388	53.2768	15.6707
0.762	10.0	7000	0.5388	53.3796	15.7012
0.7479	11.4286	8000	0.5388	53.3539	15.6751
0.8233	12.8571	9000	0.5388	53.4053	15.6794
0.7284	14.2857	10000	0.5388	53.3025	15.6925
0.7415	15.7143	11000	0.5388	53.3282	15.6751

Framework versions

• Transformers 4.46.2
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3