tgrhn/wav2vec2-turkish-6

Wav2Vec2-XLS-TR

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the Common Voice 17 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2069

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: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
4.0821	1.0	1451	3.1704
0.8611	2.0	2902	0.3721
0.4726	3.0	4353	0.3021
0.3757	4.0	5804	0.2881
0.3188	5.0	7255	0.2580
0.2837	6.0	8706	0.2529
0.2625	7.0	10157	0.2410
0.2436	8.0	11608	0.2379
0.2222	9.0	13059	0.2246
0.2193	10.0	14510	0.2355
0.2077	11.0	15961	0.2199
0.1912	12.0	17412	0.2161
0.1861	13.0	18863	0.2128
0.1732	14.0	20314	0.2164
0.1626	15.0	21765	0.2113
0.1643	16.0	23216	0.2190
0.156	17.0	24667	0.2178
0.1549	18.0	26118	0.2282
0.1391	19.0	27569	0.2014
0.1337	20.0	29020	0.2117
0.1335	21.0	30471	0.2058
0.1289	22.0	31922	0.2089
0.1211	23.0	33373	0.2114
0.1167	24.0	34824	0.2093
0.1123	25.0	36275	0.2121
0.115	26.0	37726	0.2082
0.1034	27.0	39177	0.2109
0.1053	28.0	40628	0.2064
0.1003	29.0	42079	0.2083
0.0978	30.0	43530	0.2069

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1