Modelpublic

waelhasan/wav2vec2-l-xls-r-300m-official

sourceHugging Faceapache-2.0updated 7mo agoView on Hugging Face

0likes6downloads

Model Card

wav2vec2-l-xls-r-300m-official

This model is a fine-tuned version of Khalsuu/filipino-wav2vec2-l-xls-r-300m-official on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.8308
—Cer: 0.1332

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
—trainbatchsize: 32
—evalbatchsize: 16
—seed: 42
—optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 20
—training_steps: 2000
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
7.8367	1.1905	100	3.5769	0.4891
2.2149	2.3810	200	1.1669	0.1616
1.3048	3.5714	300	0.9331	0.1494
1.0685	4.7619	400	0.8628	0.1491
0.9243	5.9524	500	0.8304	0.1341
0.7994	7.1429	600	0.8677	0.1385
0.7299	8.3333	700	0.8713	0.1352
0.6595	9.5238	800	0.8855	0.1350
0.6130	10.7143	900	0.9014	0.1340
0.5657	11.9048	1000	0.9150	0.1330
0.5385	13.0952	1100	0.9400	0.1336
0.5035	14.2857	1200	0.9942	0.1312
0.4649	15.4762	1300	1.0154	0.1295
0.4608	16.6667	1400	1.0111	0.1304
0.4340	17.8571	1500	1.0336	0.1297
0.4239	19.0476	1600	1.0370	0.1316
0.4089	20.2381	1700	1.0630	0.1315
0.3990	21.4286	1800	1.0666	0.1295
0.3989	22.6190	1900	1.0710	0.1295
0.3789	23.8095	2000	1.0815	0.1310

Framework versions

—Transformers 5.2.0
—Pytorch 2.9.0+cu126
—Datasets 4.0.0
—Tokenizers 0.22.2