Modelpublic

golesheed/wav2vec2-large-xls-r-1b-frisian-cv-13-elderly-augmented

sourceHugging Faceapache-2.0updated 3y agoView on Hugging Face

0likes31downloads

Model Card

wav2vec2-large-xls-r-1b-frisian-cv-13-elderly-augmented

This model is a fine-tuned version of greenw0lf/wav2vec2-large-xls-r-1b-frisian on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.2158
—Wer: 0.2709

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: 9e-06
—trainbatchsize: 32
—evalbatchsize: 8
—seed: 42
—optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_ratio: 0.1
—num_epochs: 40
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.7704	2.5	50	3.5242	1.0040
2.7638	5.0	100	1.8583	1.0523
1.4635	7.5	150	0.5541	0.5861
0.6753	10.0	200	0.2995	0.3353
0.4331	12.5	250	0.2439	0.2964
0.7712	15.0	300	0.2344	0.2879
0.6968	17.5	350	0.2287	0.2839
0.2765	20.0	400	0.2249	0.2776
0.4245	22.5	450	0.2202	0.2776
0.3709	25.0	500	0.2192	0.2754
0.4728	27.5	550	0.2195	0.2763
0.6297	30.0	600	0.2191	0.2736
0.5297	32.5	650	0.2186	0.2758
0.3543	35.0	700	0.2171	0.2731
0.2973	37.5	750	0.2167	0.2713
0.4558	40.0	800	0.2158	0.2709

Framework versions

—Transformers 4.27.4
—Pytorch 2.0.0+cu117
—Datasets 2.11.0
—Tokenizers 0.13.2