Edit model card

wav2vec2-bert

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the Yogera dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2157
  • Wer: 0.1291
  • Cer: 0.0296

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: 16
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 100
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Cer
0.6428 1.0 257 0.1958 0.2392 0.0488
0.1608 2.0 514 0.1623 0.1868 0.0393
0.1216 3.0 771 0.1471 0.1663 0.0368
0.1001 4.0 1028 0.1483 0.1601 0.0351
0.0859 5.0 1285 0.1471 0.1497 0.0332
0.0742 6.0 1542 0.1478 0.1468 0.0315
0.0641 7.0 1799 0.1642 0.1476 0.0326
0.0544 8.0 2056 0.1520 0.1461 0.0322
0.0489 9.0 2313 0.1596 0.1386 0.0312
0.0452 10.0 2570 0.1521 0.1408 0.0320
0.04 11.0 2827 0.1754 0.1395 0.0306
0.0371 12.0 3084 0.1703 0.1405 0.0309
0.0329 13.0 3341 0.1657 0.1447 0.0318
0.0323 14.0 3598 0.1695 0.1327 0.0298
0.0282 15.0 3855 0.1852 0.1356 0.0310
0.0237 16.0 4112 0.1728 0.1399 0.0308
0.0229 17.0 4369 0.1810 0.1301 0.0291
0.02 18.0 4626 0.1781 0.1367 0.0304
0.0204 19.0 4883 0.2039 0.1329 0.0293
0.0186 20.0 5140 0.1929 0.1366 0.0302
0.0164 21.0 5397 0.2022 0.1356 0.0301
0.0154 22.0 5654 0.1787 0.1307 0.0293
0.0127 23.0 5911 0.2086 0.1296 0.0290
0.0129 24.0 6168 0.2094 0.1281 0.0287
0.0108 25.0 6425 0.2148 0.1254 0.0280
0.0122 26.0 6682 0.2091 0.1339 0.0305
0.0106 27.0 6939 0.2030 0.1315 0.0295
0.0102 28.0 7196 0.2092 0.1241 0.0282
0.0088 29.0 7453 0.2078 0.1290 0.0287
0.008 30.0 7710 0.2112 0.1298 0.0282
0.0084 31.0 7967 0.1972 0.1305 0.0295
0.0074 32.0 8224 0.2130 0.1337 0.0293
0.0062 33.0 8481 0.2141 0.1308 0.0297
0.0065 34.0 8738 0.2151 0.1319 0.0296
0.0079 35.0 8995 0.2070 0.1253 0.0279
0.0059 36.0 9252 0.2229 0.1267 0.0285
0.0071 37.0 9509 0.2218 0.1295 0.0297
0.0066 38.0 9766 0.2157 0.1291 0.0296

Framework versions

  • Transformers 4.45.2
  • Pytorch 2.1.0+cu118
  • Datasets 3.0.1
  • Tokenizers 0.20.1
Downloads last month
5
Safetensors
Model size
606M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for asr-africa/w2v-bert-2.0-yogera-lg-26hrs-v1

Finetuned
(185)
this model

Evaluation results