w2v-bert-2.0-zulu-nchlt

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the NCHLT Dataset of Zulu Language phoen calls. It achieves the following results on the evaluation set:

  • Loss: 0.1494
  • Wer: 0.1008

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
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 32
  • optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 10
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
2.2571 0.2293 300 0.3025 0.3412
0.2601 0.4585 600 0.2271 0.2810
0.2078 0.6878 900 0.1966 0.2427
0.1806 0.9171 1200 0.1627 0.2067
0.159 1.1464 1500 0.1538 0.1956
0.1447 1.3756 1800 0.1408 0.1812
0.1365 1.6049 2100 0.1462 0.1733
0.1312 1.8342 2400 0.1376 0.1660
0.1221 2.0634 2700 0.1260 0.1611
0.1037 2.2927 3000 0.1258 0.1558
0.1023 2.5220 3300 0.1136 0.1499
0.1017 2.7512 3600 0.1106 0.1405
0.0999 2.9805 3900 0.1153 0.1454
0.0818 3.2098 4200 0.1180 0.1381
0.08 3.4391 4500 0.1098 0.1340
0.0791 3.6683 4800 0.1086 0.1340
0.0787 3.8976 5100 0.1069 0.1308
0.0694 4.1269 5400 0.1040 0.1305
0.0629 4.3561 5700 0.1053 0.1311
0.0642 4.5854 6000 0.1064 0.1312
0.0593 4.8147 6300 0.1008 0.1233
0.0599 5.0439 6600 0.0974 0.1204
0.0438 5.2732 6900 0.1009 0.1168
0.0465 5.5025 7200 0.0959 0.1168
0.0486 5.7318 7500 0.1019 0.1175
0.0474 5.9610 7800 0.0973 0.1123
0.0338 6.1903 8100 0.1011 0.1140
0.032 6.4196 8400 0.1165 0.1131
0.0324 6.6488 8700 0.1062 0.1132
0.0303 6.8781 9000 0.1032 0.1132
0.0265 7.1074 9300 0.1233 0.1097
0.0204 7.3366 9600 0.1147 0.1134
0.0193 7.5659 9900 0.1152 0.1055
0.0192 7.7952 10200 0.1122 0.1067
0.0176 8.0245 10500 0.1241 0.1067
0.0098 8.2537 10800 0.1188 0.1061
0.0089 8.4861 11100 0.1304 0.1021
0.0079 8.7153 11400 0.1346 0.1008
0.0075 8.9446 11700 0.1361 0.1034
0.0055 9.1739 12000 0.1472 0.1047
0.0046 9.4031 12300 0.1467 0.1008
0.0047 9.6324 12600 0.1472 0.1016
0.0045 9.8617 12900 0.1494 0.1008

Framework versions

  • Transformers 4.46.2
  • Pytorch 2.4.1+cu121
  • Datasets 2.18.0
  • Tokenizers 0.20.0
Downloads last month
6
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 aconeil/w2v-bert-2.0-zulu-nchlt

Finetuned
(238)
this model