kiranpantha/wav2vec2-large-xls-r-300m-nepali

XLSR-300M-Nepali

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the OpenSLR54 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2681
• Wer: 0.5244

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.2642	0.0722	300	2.9627	1.0
2.1949	0.1444	600	1.5526	1.0160
1.4595	0.2166	900	1.1674	0.9810
1.2128	0.2888	1200	0.9901	0.9668
0.976	0.3610	1500	0.6942	0.7696
0.8267	0.4332	1800	0.6314	0.7552
0.7542	0.5054	2100	0.5522	0.7156
0.7228	0.5776	2400	0.5210	0.6960
0.6707	0.6498	2700	0.4744	0.6581
0.6368	0.7220	3000	0.4529	0.6535
0.5944	0.7942	3300	0.4229	0.6264
0.5651	0.8664	3600	0.4061	0.6161
0.5469	0.9386	3900	0.3788	0.6103
0.5308	1.0108	4200	0.3668	0.5957
0.4684	1.0830	4500	0.3509	0.5920
0.4382	1.1552	4800	0.3398	0.5920
0.4424	1.2274	5100	0.3260	0.5767
0.4159	1.2996	5400	0.3189	0.5690
0.419	1.3718	5700	0.3067	0.5581
0.4114	1.4440	6000	0.3019	0.5568
0.3903	1.5162	6300	0.2982	0.5549
0.3915	1.5884	6600	0.2887	0.5493
0.3789	1.6606	6900	0.2813	0.5398
0.3725	1.7329	7200	0.2763	0.5339
0.3706	1.8051	7500	0.2704	0.5285
0.3624	1.8773	7800	0.2706	0.5264
0.357	1.9495	8100	0.2681	0.5244

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1