wav2vec2-xls-r-300m-kab-30m

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: 1.4897
• Wer: 0.7494
• Cer: 0.2461

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
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.6174	16.1290	500	3.0994	1.0	1.0
3.0584	32.2581	1000	3.0722	1.0	1.0
1.8588	48.3871	1500	1.7183	0.9851	0.5910
1.0303	64.5161	2000	1.3911	0.8739	0.3154
0.7667	80.6452	2500	1.3397	0.8308	0.2813
0.5824	96.7742	3000	1.4542	0.8056	0.2703
0.4857	112.9032	3500	1.4650	0.8039	0.2636
0.445	129.0323	4000	1.5069	0.7729	0.2557
0.4163	145.1613	4500	1.4767	0.7540	0.2477
0.3967	161.2903	5000	1.4897	0.7494	0.2461

Framework versions

• Transformers 4.50.0
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1