Modelpublic

alinerodrigues/wav2vec2-xlsr-1b-mecita-portuguese-all-clean-03

sourceHugging Faceapache-2.0updated 3y agoView on Hugging Face

0likes11downloads

Model Card

wav2vec2-xlsr-1b-mecita-portuguese-all-clean-03

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

—Loss: 0.1307
—Wer: 0.0783
—Cer: 0.0231

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: 3e-05
—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
—num_epochs: 100
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
21.675	1.0	67	4.4178	0.9636	0.9162
6.0023	2.0	134	3.9535	0.9818	0.9783
4.2873	3.0	201	3.7663	0.9719	0.9834
4.2873	4.0	268	3.0601	0.9881	0.9680
3.5535	5.0	335	0.4943	0.5511	0.1220
1.0258	6.0	402	0.1872	0.1150	0.0329
1.0258	7.0	469	0.1534	0.0998	0.0283
0.3109	8.0	536	0.1406	0.0899	0.0267
0.2534	9.0	603	0.1432	0.0774	0.0237
0.2534	10.0	670	0.1325	0.08	0.0237
0.2109	11.0	737	0.1342	0.0803	0.0233
0.1998	12.0	804	0.1322	0.0856	0.0245
0.1998	13.0	871	0.1355	0.0807	0.0233
0.1696	14.0	938	0.1343	0.0754	0.0224
0.1875	15.0	1005	0.1320	0.0747	0.0232
0.1875	16.0	1072	0.1361	0.0737	0.0226
0.1501	17.0	1139	0.1307	0.0783	0.0231
0.1433	18.0	1206	0.1410	0.0790	0.0242
0.1433	19.0	1273	0.1412	0.0840	0.0246
0.1336	20.0	1340	0.1320	0.0740	0.0230
0.1443	21.0	1407	0.1308	0.0754	0.0228
0.1443	22.0	1474	0.1366	0.0750	0.0224
0.1253	23.0	1541	0.1396	0.0793	0.0237
0.1145	24.0	1608	0.1317	0.0754	0.0230
0.1145	25.0	1675	0.1346	0.0790	0.0241
0.1071	26.0	1742	0.1343	0.0724	0.0232
0.0989	27.0	1809	0.1343	0.0750	0.0232
0.0989	28.0	1876	0.1332	0.0744	0.0228
0.0982	29.0	1943	0.1348	0.0714	0.0216
0.0934	30.0	2010	0.1351	0.0698	0.0219
0.0934	31.0	2077	0.1349	0.0740	0.0231
0.088	32.0	2144	0.1341	0.0691	0.0225
0.086	33.0	2211	0.1336	0.0731	0.0227
0.086	34.0	2278	0.1398	0.0757	0.0240
0.0907	35.0	2345	0.1494	0.0714	0.0225
0.0974	36.0	2412	0.1455	0.0754	0.0232
0.0974	37.0	2479	0.1436	0.0724	0.0237

Framework versions

—Transformers 4.28.0
—Pytorch 2.2.1+cu121
—Datasets 2.17.0
—Tokenizers 0.13.3