wav2vec2-xls-r-2b-5e-sw-asr

This model is a fine-tuned version of facebook/wav2vec2-xls-r-2b on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: nan
• Wer: 0.9999

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.7865	0.2179	400	inf	0.7489
0.599	0.4357	800	inf	0.7282
0.5505	0.6536	1200	inf	0.6650
0.5238	0.8715	1600	inf	0.7014
0.4933	1.0893	2000	inf	0.5907
0.4488	1.3072	2400	inf	0.5622
0.4442	1.5251	2800	inf	0.5664
0.4353	1.7429	3200	inf	0.5418
0.4059	1.9608	3600	inf	0.5334
0.3737	2.1786	4000	inf	0.5037
0.3547	2.3965	4400	inf	0.5097
0.3594	2.6144	4800	inf	0.5031
0.563	2.8322	5200	inf	0.8574
1.4779	3.0501	5600	inf	0.9908
2.4352	3.2680	6000	nan	0.9946
214.8165	3.4858	6400	nan	0.9998
0.0	3.7037	6800	nan	0.9996
0.0	3.9216	7200	nan	0.9999
0.0	4.1394	7600	nan	0.9997
0.0	4.3573	8000	nan	0.9997
0.0	4.5752	8400	nan	0.9996
0.0	4.7930	8800	nan	0.9999

Framework versions

• Transformers 4.56.2
• Pytorch 2.8.0+cu126
• Datasets 3.6.0
• Tokenizers 0.22.0