Modelpublic

jfealko/wav2vec2-large-xls-r-300m-irish-local

sourceHugging Faceapache-2.0updated 5y agoView on Hugging Face

0likes65downloads

Model Card

wav2vec2-large-xls-r-300m-irish-local

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

—Loss: 2.0788
—Wer: 0.7527

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: 50
—num_epochs: 90
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
8.3839	2.94	50	3.3021	1.0
3.0703	5.88	100	3.1749	1.0
3.1744	8.82	150	3.0452	1.0
2.9719	11.76	200	2.9767	1.0
2.9539	14.71	250	2.9992	1.0
2.9438	17.65	300	2.9767	1.0
2.9296	20.59	350	2.9475	1.0
2.9269	23.53	400	2.9402	1.0
2.9116	26.47	450	2.9255	1.0
2.8326	29.41	500	2.7238	1.0
2.5758	32.35	550	2.3599	0.9900
2.1242	35.29	600	1.8478	0.9491
1.4603	38.24	650	1.5991	0.9002
1.0287	41.18	700	1.5931	0.8434
0.7687	44.12	750	1.6493	0.8253
0.571	47.06	800	1.6889	0.8057
0.4598	50.0	850	1.7521	0.7978
0.3902	52.94	900	1.9074	0.7975
0.318	55.88	950	1.9352	0.8133
0.3026	58.82	1000	2.0157	0.8028
0.2862	61.76	1050	1.9231	0.7720
0.2696	64.71	1100	1.9256	0.7644
0.2528	67.65	1150	2.0277	0.7741
0.2051	70.59	1200	1.9921	0.7550
0.2018	73.53	1250	2.0416	0.7615
0.187	76.47	1300	2.0861	0.7635
0.1749	79.41	1350	2.0926	0.7577
0.1713	82.35	1400	2.0632	0.7533
0.1518	85.29	1450	2.0903	0.7542
0.16	88.24	1500	2.0788	0.7527

Framework versions

—Transformers 4.11.3
—Pytorch 1.10.0+cu113
—Datasets 1.18.3
—Tokenizers 0.10.3