Modelpublic

VicentCano/toki-pona-wav2vec2-xls-r-300m-common-voice-17

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes181downloads

Model Card

tokiponatestcognwav2vec2-xls-r-300m

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the commonvoice17_0 dataset. It achieves the following results on the evaluation set:

—Loss: 0.0718
—Wer: 0.0457

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
—trainbatchsize: 4
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 8
—totaltrainbatch_size: 32
—optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—num_epochs: 14
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.9966	147	2.6699	1.0
No log	2.0	295	0.6492	0.5195
No log	2.9966	442	0.1896	0.1911
2.6379	4.0	590	0.1464	0.1274
2.6379	4.9966	737	0.1177	0.1073
2.6379	6.0	885	0.1089	0.0935
0.1404	6.9966	1032	0.0978	0.0792
0.1404	8.0	1180	0.0957	0.0735
0.1404	8.9966	1327	0.0888	0.0697
0.1404	10.0	1475	0.0741	0.0584
0.0683	10.9966	1622	0.0710	0.0519
0.0683	12.0	1770	0.0687	0.0477
0.0683	12.9966	1917	0.0709	0.0458
0.0383	13.9525	2058	0.0718	0.0457

Framework versions

—Transformers 4.46.3
—Pytorch 2.5.1+cu121
—Datasets 3.1.0
—Tokenizers 0.20.3