wav2vec2-xls-r-300m-kab

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: 1.1576
• Wer: 0.8033

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: 7e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.98) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 80
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.6735	4.7170	250	3.1798	1.0
3.1693	9.4340	500	3.0853	1.0
3.1029	14.1509	750	3.0879	1.0
3.0558	18.8679	1000	3.0586	1.0
2.9616	23.5849	1250	2.9119	1.0
2.5156	28.3019	1500	2.3047	1.0
1.9201	33.0189	1750	1.6438	0.9931
1.5374	37.7358	2000	1.3795	0.9524
1.2811	42.4528	2250	1.2664	0.9111
1.1301	47.1698	2500	1.2380	0.8899
1.0473	51.8868	2750	1.2092	0.8675
0.9705	56.6038	3000	1.1904	0.8463
0.9097	61.3208	3250	1.1488	0.8320
0.8127	66.0377	3500	1.1410	0.8200
0.8294	70.7547	3750	1.1527	0.8050
0.8403	75.4717	4000	1.1576	0.8033

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1