wav2vec2-burak-new-300-v2

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

• Loss: 0.7018
• Wer: 0.3641

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: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 141

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.2967	8.62	500	1.0561	0.8351
0.5199	17.24	1000	0.6019	0.5054
0.2249	25.86	1500	0.6036	0.4576
0.1573	34.48	2000	0.6680	0.4532
0.1301	43.1	2500	0.6823	0.4396
0.1108	51.72	3000	0.6630	0.4263
0.0941	60.34	3500	0.6574	0.4226
0.0797	68.97	4000	0.6796	0.4174
0.0689	77.59	4500	0.6426	0.4088
0.0612	86.21	5000	0.6615	0.3949
0.0528	94.83	5500	0.6769	0.3906
0.0515	103.45	6000	0.6073	0.3827
0.0394	112.07	6500	0.7099	0.3767
0.0352	120.69	7000	0.7082	0.3688
0.0324	129.31	7500	0.6939	0.3699
0.028	137.93	8000	0.7018	0.3641

Framework versions

• Transformers 4.21.3
• Pytorch 1.12.1+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1