mdd-arabic-e5-xlsr1b-specaugment-beam

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: 1.2205
• Per: 0.2245
• Accuracy: 0.7621
• Precision: 0.2869
• Recall: 0.2746
• F1 Macro: 0.2804

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: 16
• 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
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Accuracy	F1 Macro	Validation Loss	Per	Precision	Recall
2.1605	1.0	728	0.7864	0.2971	0.7895	0.1988	0.2976	0.3045
1.0854	2.0	1456	0.7764	0.2950	0.9990	0.2084	0.3008	0.2896
0.7894	3.0	2184	0.7773	0.2879	1.0016	0.2078	0.2959	0.2804
0.6151	4.0	2912	0.7765	0.2978	1.0822	0.2104	0.3033	0.2927
0.6811	5.0	3640	0.7697	0.2807	1.0894	0.2127	0.2847	0.2778
0.5452	6.0	4368	0.7715	0.2888	1.0083	0.2130	0.2928	0.2857
0.4816	7.0	5096	1.1389	0.2206	0.7654	0.2902	0.2862	0.2874
0.4420	8.0	5824	1.2157	0.2247	0.7617	0.2824	0.2741	0.2777
0.4235	9.0	6552	1.1786	0.2312	0.7564	0.2765	0.2693	0.2721
0.3895	10.0	7280	1.2156	0.2285	0.7600	0.2879	0.2785	0.2826
0.5042	11.0	8008	1.2205	0.2245	0.7621	0.2869	0.2746	0.2804

Framework versions

• Transformers 5.15.0
• Pytorch 2.13.0+cu130
• Datasets 5.0.1
• Tokenizers 0.22.2