You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

w2vbert-luganda-waxal

This model is a fine-tuned version of sulaimank/w2v-bert-2.0-lg-CV-Fleurs-300 on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.1126
  • Wer: 0.1278
  • Cer: 0.0285
  • Zindi: 0.9219

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: 0.0001
  • train_batch_size: 16
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 64
  • optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 60.0

Training results

Training Loss Epoch Step Validation Loss Wer Cer Zindi
0.2041 0.1116 500 0.1647 0.1696 0.0387 0.8958
0.2117 0.2233 1000 0.1667 0.1687 0.0407 0.8953
0.2281 0.3349 1500 0.1568 0.1708 0.0410 0.8941
0.2158 0.4466 2000 0.1530 0.1697 0.0370 0.8966
0.2199 0.5582 2500 0.1358 0.1515 0.0337 0.9074
0.2099 0.6699 3000 0.1419 0.1677 0.0370 0.8976
0.2347 0.7815 3500 0.1400 0.1505 0.0337 0.9079
0.2444 0.8932 4000 0.1436 0.1570 0.0358 0.9036
0.2041 1.0047 4500 0.1294 0.1497 0.0330 0.9087
0.1984 1.1163 5000 0.1346 0.1461 0.0324 0.9108
0.1860 1.2280 5500 0.1340 0.1488 0.0336 0.9088
0.1821 1.3396 6000 0.1289 0.1421 0.0318 0.9130
0.2105 1.4513 6500 0.1321 0.1494 0.0330 0.9088
0.1951 1.5629 7000 0.1290 0.1441 0.0326 0.9116
0.1906 1.6746 7500 0.1339 0.1450 0.0328 0.9111
0.1824 1.7862 8000 0.1285 0.1401 0.0323 0.9138
0.2062 1.8978 8500 0.1271 0.1387 0.0326 0.9144
0.1450 2.0094 9000 0.1205 0.1362 0.0304 0.9167
0.1598 2.1210 9500 0.1225 0.1377 0.0312 0.9155
0.1942 2.2327 10000 0.1201 0.1386 0.0307 0.9154
0.1446 2.3443 10500 0.1236 0.1377 0.0314 0.9155
0.1635 2.4560 11000 0.1154 0.1329 0.0297 0.9187
0.1630 2.5676 11500 0.1194 0.1350 0.0296 0.9177
0.1805 2.6792 12000 0.1215 0.1412 0.0319 0.9134
0.1561 2.7909 12500 0.1155 0.1313 0.0292 0.9198
0.1740 2.9025 13000 0.1202 0.1244 0.0288 0.9234
0.1348 3.0141 13500 0.1191 0.1286 0.0291 0.9212
0.1291 3.1257 14000 0.1207 0.1325 0.0294 0.9191
0.1252 3.2374 14500 0.1174 0.1332 0.0297 0.9186
0.1305 3.3490 15000 0.1167 0.1353 0.0298 0.9175
0.1363 3.4606 15500 0.1163 0.1314 0.0290 0.9198
0.1415 3.5723 16000 0.1145 0.1314 0.0296 0.9195
0.1504 3.6839 16500 0.1141 0.1267 0.0283 0.9225
0.1669 3.7956 17000 0.1142 0.1276 0.0285 0.9220
0.1493 3.9072 17500 0.1126 0.1233 0.0275 0.9246
0.1197 4.0188 18000 0.1134 0.1241 0.0279 0.9240
0.1110 4.1304 18500 0.1123 0.1290 0.0287 0.9212
0.1084 4.2420 19000 0.1131 0.1211 0.0273 0.9258
0.1197 4.3537 19500 0.1114 0.1230 0.0278 0.9246
0.1393 4.4653 20000 0.1151 0.1302 0.0293 0.9203
0.1318 4.5770 20500 0.1111 0.1207 0.0275 0.9259
0.1325 4.6886 21000 0.1100 0.1231 0.0271 0.9249
0.1249 4.8003 21500 0.1137 0.1210 0.0269 0.9260
0.1406 4.9119 22000 0.1117 0.1252 0.0283 0.9232
0.0883 5.0234 22500 0.1093 0.1181 0.0271 0.9274
0.1049 5.1351 23000 0.1107 0.1212 0.0270 0.9259
0.0997 5.2467 23500 0.1123 0.1277 0.0286 0.9218
0.1148 5.3584 24000 0.1147 0.1325 0.0306 0.9184
0.1053 5.4700 24500 0.1090 0.1223 0.0275 0.9251
0.1019 5.5817 25000 0.1127 0.1259 0.0284 0.9228
0.1110 5.6933 25500 0.1100 0.1184 0.0271 0.9272
0.0972 5.8050 26000 0.1059 0.1203 0.0269 0.9264
0.1204 5.9166 26500 0.1152 0.1260 0.0280 0.9230
0.0748 6.0281 27000 0.1088 0.1219 0.0272 0.9254
0.0747 6.1398 27500 0.1126 0.1278 0.0285 0.9219

Framework versions

  • Transformers 5.13.0
  • Pytorch 2.12.1+cu130
  • Datasets 3.6.0
  • Tokenizers 0.22.2
Downloads last month
393
Safetensors
Model size
0.6B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for sulaimank/w2vbert-luganda-waxal

Finetuned
(2)
this model
Finetunes
1 model