Edit model card

hing-mbert-finetuned-non-code-mixed-DS

This model is a fine-tuned version of l3cube-pune/hing-mbert on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 1.6572
  • Accuracy: 0.6429
  • Precision: 0.6334
  • Recall: 0.6231
  • F1: 0.6262

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: 4.932923543227153e-05
  • train_batch_size: 4
  • eval_batch_size: 8
  • seed: 43
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 4

Training results

Training Loss Epoch Step Validation Loss Accuracy Precision Recall F1
1.005 0.5 926 0.9346 0.5707 0.5844 0.5274 0.5108
0.969 1.0 1852 1.0295 0.5858 0.5893 0.5685 0.5455
0.8976 1.5 2778 1.0491 0.5739 0.5712 0.5295 0.5152
0.8578 2.0 3704 0.8577 0.6343 0.6443 0.6379 0.6318
0.7164 2.5 4630 1.3325 0.6300 0.6219 0.5932 0.5939
0.7391 3.0 5556 1.0329 0.6537 0.6489 0.6519 0.6467
0.5454 3.5 6482 1.6031 0.6375 0.6525 0.6188 0.6218
0.4927 4.0 7408 1.6572 0.6429 0.6334 0.6231 0.6262

Framework versions

  • Transformers 4.21.3
  • Pytorch 1.12.1+cu113
  • Datasets 2.4.0
  • Tokenizers 0.12.1
Downloads last month
4