Edit model card

deberta-v3-base-cotat

This model is a fine-tuned version of microsoft/deberta-v3-base on the DandinPower/review_cleanonlytitleandtext dataset. It achieves the following results on the evaluation set:

  • Loss: 1.4985
  • Accuracy: 0.623
  • Macro F1: 0.6247

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: 4.5e-05
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 1500
  • num_epochs: 5

Training results

Training Loss Epoch Step Validation Loss Accuracy Macro F1
1.0223 0.14 500 0.9610 0.592 0.5971
1.0108 0.29 1000 0.9378 0.6044 0.6083
0.9323 0.43 1500 0.9605 0.589 0.5652
0.9651 0.57 2000 0.9845 0.5797 0.5687
0.928 0.71 2500 0.9521 0.5907 0.5656
0.9205 0.86 3000 0.9073 0.603 0.5740
0.9243 1.0 3500 0.8876 0.616 0.6113
0.8545 1.14 4000 0.8631 0.6267 0.6290
0.8267 1.29 4500 0.8908 0.624 0.6185
0.8175 1.43 5000 0.8771 0.6173 0.6222
0.8613 1.57 5500 0.9564 0.6209 0.6081
0.8138 1.71 6000 0.9246 0.6089 0.6063
0.7314 1.86 6500 0.9030 0.6329 0.6313
0.8287 2.0 7000 0.8753 0.6211 0.6235
0.6963 2.14 7500 0.9700 0.6247 0.6257
0.7034 2.29 8000 0.9592 0.6234 0.6220
0.679 2.43 8500 0.8994 0.6233 0.6272
0.7207 2.57 9000 1.0013 0.6236 0.6183
0.6992 2.71 9500 0.9385 0.6169 0.6219
0.7032 2.86 10000 0.9247 0.6366 0.6364
0.6949 3.0 10500 0.9615 0.6239 0.6281
0.5581 3.14 11000 1.0439 0.6217 0.6267
0.55 3.29 11500 1.1205 0.6259 0.6232
0.5496 3.43 12000 1.1122 0.6226 0.6267
0.5462 3.57 12500 1.0692 0.6251 0.6263
0.5121 3.71 13000 1.1563 0.6197 0.6214
0.531 3.86 13500 1.1123 0.6261 0.6256
0.5256 4.0 14000 1.1194 0.6247 0.6264
0.3908 4.14 14500 1.3631 0.6204 0.6210
0.4439 4.29 15000 1.4810 0.6204 0.6211
0.4252 4.43 15500 1.4454 0.6211 0.6217
0.3721 4.57 16000 1.5315 0.6204 0.6231
0.369 4.71 16500 1.4797 0.6184 0.6190
0.3907 4.86 17000 1.4857 0.6219 0.6234
0.4022 5.0 17500 1.4985 0.623 0.6247

Framework versions

  • Transformers 4.39.3
  • Pytorch 2.2.2+cu121
  • Datasets 2.18.0
  • Tokenizers 0.15.2
Downloads last month
3
Safetensors
Model size
184M params
Tensor type
F32
·
Inference API
This model can be loaded on Inference API (serverless).

Finetuned from

Dataset used to train DandinPower/deberta-v3-base-cotat

Collection including DandinPower/deberta-v3-base-cotat

Evaluation results

  • Accuracy on DandinPower/review_cleanonlytitleandtext
    self-reported
    0.623