Modelpublic

Samuael/geez_t5-15k

sourceHugging Faceupdated 2y agoView on Hugging Face

0likes10downloads

Model Card

geez_t5-15k

This model is a fine-tuned version of Samuael/geez_t5-15k on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 1.6541
—Wer: 0.2404
—Cer: 0.0668
—Bleu: 54.5297

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: 0.0002
—trainbatchsize: 64
—evalbatchsize: 128
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—num_epochs: 10
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Bleu
0.0382	1.0	313	1.4979	0.2436	0.0667	55.2918
0.0411	2.0	626	1.5980	0.2419	0.0653	55.4560
0.0622	3.0	939	1.5221	0.2383	0.0683	55.5429
0.0505	4.0	1252	1.5719	0.2388	0.0662	55.6735
0.0517	5.0	1565	1.6251	0.2373	0.0666	55.7046
0.0396	6.0	1878	1.5978	0.2431	0.0680	54.5671
0.0512	7.0	2191	1.5677	0.2675	0.1019	54.3154
0.0655	8.0	2504	1.5891	0.2448	0.0676	54.6644
0.05	9.0	2817	1.5952	0.2470	0.0703	54.7796
0.0415	10.0	3130	1.6541	0.2404	0.0668	54.5297

Framework versions

—Transformers 4.38.2
—Pytorch 2.2.1+cu121
—Datasets 2.18.0
—Tokenizers 0.15.2