wav2vec2-turkish-300m-4

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the fleurs dataset. It achieves the following results on the evaluation set:

• Loss: 0.2826
• Wer: 0.2126

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: 4
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 0.1
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.5303	0.6983	500	0.5432	0.5985
0.6272	1.3966	1000	0.3629	0.4617
0.415	2.0950	1500	0.2861	0.3580
0.2507	2.7933	2000	0.2692	0.3357
0.1909	3.4916	2500	0.2619	0.3009
0.1731	4.1899	3000	0.2843	0.3085
0.1358	4.8883	3500	0.2678	0.2752
0.1002	5.5866	4000	0.2623	0.2667
0.0915	6.2849	4500	0.2551	0.2515
0.0749	6.9832	5000	0.2579	0.2423
0.0695	7.6816	5500	0.2680	0.2316
0.0499	8.3799	6000	0.2774	0.2261
0.0464	9.0782	6500	0.2793	0.2179
0.0398	9.7765	7000	0.2826	0.2126

Framework versions

• Transformers 4.40.0
• Pytorch 2.2.2+cu121
• Datasets 2.17.1
• Tokenizers 0.19.1