Modelpublic

nrshoudi/huber_arabic_mdd_v2

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes10downloads

Model Card

huberarabicmdd_v2

This model is a fine-tuned version of facebook/hubert-large-ll60k on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.2858
—Wer: 0.0564
—Cer: 0.0459

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: 16
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 32
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—num_epochs: 20
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.3219	0.9951	102	3.2861	1.0	1.0
3.2152	2.0	205	3.1685	1.0	1.0
2.5507	2.9951	307	2.3708	0.9718	0.9819
0.5766	4.0	410	0.6351	0.2216	0.2046
0.2255	4.9951	512	0.3469	0.0889	0.0740
0.1148	6.0	615	0.3393	0.0776	0.0635
0.1222	6.9951	717	0.3368	0.0688	0.0535
0.075	8.0	820	0.2846	0.0610	0.0479
0.0631	8.9951	922	0.2948	0.0589	0.0453
0.0365	10.0	1025	0.2657	0.0552	0.0432
0.0484	10.9951	1127	0.2631	0.0573	0.0458
0.046	12.0	1230	0.2817	0.0572	0.0462
0.0326	12.9951	1332	0.2807	0.0587	0.0473
0.0379	14.0	1435	0.2682	0.0590	0.0479
0.0328	14.9951	1537	0.2773	0.0545	0.0440
0.0398	16.0	1640	0.2727	0.0576	0.0462
0.0165	16.9951	1742	0.2844	0.0573	0.0466
0.0201	18.0	1845	0.2812	0.0564	0.0455
0.0207	18.9951	1947	0.2860	0.0569	0.0465
0.0194	19.9024	2040	0.2858	0.0564	0.0459

Framework versions

—Transformers 4.40.0
—Pytorch 2.2.1+cu121
—Datasets 2.19.0
—Tokenizers 0.19.1