Modelpublic

csikasote/SSA-HuBERT-XL-60k-nyagen-original-train-ft-62

sourceHugging Facecc-by-nc-4.0updated 8d agoView on Hugging Face

0likes49downloads

Model Card

SSA-HuBERT-XL-60k-nyagen-original-train-ft-62

This model is a fine-tuned version of Orange/SSA-HuBERT-XL-60k on the NYAGEN - NYA dataset. It achieves the following results on the evaluation set:

—Loss: 0.3859
—Cer: 0.1442

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: 8
—evalbatchsize: 4
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 16
—optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—num_epochs: 30.0
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
7.4244	0.1988	100	2.8955	1.0
2.0195	0.3976	200	0.8332	0.2266
0.4282	0.5964	300	0.6307	0.1953
0.3053	0.7952	400	0.5550	0.1710
0.2546	0.9940	500	0.5263	0.1618
0.207	1.1928	600	0.5077	0.1573
0.1996	1.3917	700	0.4677	0.1530
0.1656	1.5905	800	0.4580	0.1508
0.1657	1.7893	900	0.4107	0.1498
0.1701	1.9881	1000	0.3960	0.1451
0.1343	2.1869	1100	0.4298	0.1426
0.1138	2.3857	1200	0.3877	0.1439
0.1229	2.5845	1300	0.3794	0.1412
0.1217	2.7833	1400	0.3874	0.1416
0.1142	2.9821	1500	0.3680	0.1436
0.0899	3.1809	1600	0.4069	0.1392
0.0914	3.3797	1700	0.4577	0.1388
0.0957	3.5785	1800	0.4039	0.1393

Framework versions

—Transformers 4.53.0
—Pytorch 2.10.0+cu128
—Datasets 4.5.0
—Tokenizers 0.21.4