wav2vec-xlsr-53-turkish-v2

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_22_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2390
• Wer: 0.3223

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: 3e-05
• train_batch_size: 16
• eval_batch_size: 8
• 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
• lr_scheduler_warmup_steps: 1000
• num_epochs: 12
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.2252	0.4601	750	3.2062	1.0
2.0823	0.9202	1500	1.7292	0.9959
0.8412	1.3804	2250	0.5604	0.6038
0.649	1.8405	3000	0.4336	0.5132
0.5598	2.3006	3750	0.3846	0.4728
0.4887	2.7607	4500	0.3499	0.4403
0.441	3.2209	5250	0.3333	0.4255
0.4429	3.6810	6000	0.3075	0.4073
0.3908	4.1411	6750	0.2941	0.3943
0.37	4.6012	7500	0.2894	0.3844
0.3651	5.0613	8250	0.2809	0.3763
0.3725	5.5215	9000	0.2738	0.3673
0.3443	5.9816	9750	0.2698	0.3619
0.327	6.4417	10500	0.2624	0.3541
0.3223	6.9018	11250	0.2579	0.3489
0.3355	7.3620	12000	0.2496	0.3420
0.3335	7.8221	12750	0.2533	0.3413
0.3242	8.2822	13500	0.2493	0.3374
0.3326	8.7423	14250	0.2454	0.3336
0.2957	9.2025	15000	0.2475	0.3314
0.281	9.6626	15750	0.2437	0.3284
0.29	10.1227	16500	0.2420	0.3268
0.293	10.5828	17250	0.2396	0.3243
0.2793	11.0429	18000	0.2406	0.3251
0.2834	11.5031	18750	0.2405	0.3232
0.2912	11.9632	19500	0.2390	0.3223

Framework versions

• Transformers 4.56.1
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.22.0