wav2vec2-turkish-300m-6

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: 0.2805
• Wer: 0.1808

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: 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: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.7044	0.6983	500	1.2056	0.8772
1.1804	1.3966	1000	0.3788	0.4238
0.5051	2.0950	1500	0.2715	0.3205
0.2605	2.7933	2000	0.2432	0.2935
0.1976	3.4916	2500	0.2535	0.2711
0.1759	4.1899	3000	0.2534	0.2589
0.1379	4.8883	3500	0.2307	0.2519
0.1045	5.5866	4000	0.2257	0.2335
0.1028	6.2849	4500	0.2408	0.2346
0.0844	6.9832	5000	0.2344	0.2285
0.0842	7.6816	5500	0.2439	0.2265
0.0685	8.3799	6000	0.2547	0.2260
0.0646	9.0782	6500	0.2509	0.2186
0.0565	9.7765	7000	0.2488	0.2146
0.0562	10.4749	7500	0.2513	0.2150
0.0479	11.1732	8000	0.2531	0.2131
0.0433	11.8715	8500	0.2636	0.2050
0.0442	12.5698	9000	0.2602	0.1959
0.0407	13.2682	9500	0.2721	0.2005
0.0378	13.9665	10000	0.2641	0.1965
0.0365	14.6648	10500	0.2715	0.1928
0.0349	15.3631	11000	0.2727	0.1924
0.0323	16.0615	11500	0.2756	0.1913
0.0299	16.7598	12000	0.2774	0.1857
0.0286	17.4581	12500	0.2701	0.1848
0.0277	18.1564	13000	0.2787	0.1841
0.0263	18.8547	13500	0.2761	0.1812
0.0241	19.5531	14000	0.2805	0.1808

Framework versions

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