wav2vec2-burak-new-300-v2-2

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: 0.6158
• Wer: 0.3094

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: 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: 241

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.5201	8.62	500	3.1581	1.0
2.1532	17.24	1000	0.6883	0.5979
0.5465	25.86	1500	0.5028	0.4432
0.3287	34.48	2000	0.4986	0.4024
0.2571	43.1	2500	0.4920	0.3824
0.217	51.72	3000	0.5265	0.3724
0.1848	60.34	3500	0.5539	0.3714
0.1605	68.97	4000	0.5689	0.3670
0.1413	77.59	4500	0.5962	0.3501
0.1316	86.21	5000	0.5732	0.3494
0.1168	94.83	5500	0.5912	0.3461
0.1193	103.45	6000	0.5766	0.3378
0.0996	112.07	6500	0.5818	0.3403
0.0941	120.69	7000	0.5986	0.3315
0.0912	129.31	7500	0.5802	0.3280
0.0865	137.93	8000	0.5878	0.3290
0.0804	146.55	8500	0.5784	0.3228
0.0739	155.17	9000	0.5791	0.3180
0.0718	163.79	9500	0.5864	0.3146
0.0681	172.41	10000	0.6104	0.3178
0.0688	181.03	10500	0.5983	0.3160
0.0657	189.66	11000	0.6228	0.3203
0.0598	198.28	11500	0.6057	0.3122
0.0597	206.9	12000	0.6094	0.3129
0.0551	215.52	12500	0.6114	0.3127
0.0507	224.14	13000	0.6056	0.3094
0.0554	232.76	13500	0.6158	0.3094

Framework versions

• Transformers 4.22.1
• Pytorch 1.12.1+cu113
• Datasets 2.5.1
• Tokenizers 0.12.1