turkish-hs-4class-prediction

This model is a fine-tuned version of dbmdz/bert-base-turkish-uncased on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.6299
  • Accuracy: 0.8386
  • Macro F1: 0.7423

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-06
  • train_batch_size: 16
  • eval_batch_size: 20
  • seed: 42
  • optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • lr_scheduler_type: linear
  • num_epochs: 10

Training results

Training Loss Epoch Step Validation Loss Accuracy Macro F1
0.9672 0.1546 100 0.7851 0.7256 0.4082
0.723 0.3091 200 0.6540 0.7633 0.5104
0.6837 0.4637 300 0.6314 0.7546 0.5525
0.617 0.6182 400 0.5643 0.7865 0.5749
0.5646 0.7728 500 0.5438 0.7961 0.5855
0.5512 0.9274 600 0.5321 0.8029 0.5996
0.5424 1.0819 700 0.5225 0.8077 0.6486
0.4391 1.2365 800 0.5219 0.8077 0.6325
0.4898 1.3910 900 0.5025 0.8164 0.6495
0.4858 1.5456 1000 0.4951 0.8232 0.6762
0.4845 1.7002 1100 0.4921 0.8261 0.6952
0.4459 1.8547 1200 0.4736 0.8425 0.7337
0.4382 2.0093 1300 0.4817 0.8087 0.6545
0.4137 2.1638 1400 0.4742 0.8357 0.7303
0.3781 2.3184 1500 0.4939 0.8213 0.6920
0.4351 2.4730 1600 0.4945 0.8309 0.7283
0.4155 2.6275 1700 0.4806 0.8367 0.7150
0.3674 2.7821 1800 0.4797 0.8357 0.7203
0.3878 2.9366 1900 0.4717 0.8406 0.7355
0.3613 3.0912 2000 0.4934 0.8367 0.7222
0.3289 3.2457 2100 0.4841 0.8377 0.7210
0.3332 3.4003 2200 0.4928 0.8386 0.7343
0.3724 3.5549 2300 0.4890 0.8386 0.7276
0.3421 3.7094 2400 0.4942 0.8425 0.7397
0.3289 3.8640 2500 0.4912 0.8454 0.7453
0.3374 4.0185 2600 0.4880 0.8454 0.7479
0.3045 4.1731 2700 0.4936 0.8415 0.7412
0.2781 4.3277 2800 0.4889 0.8444 0.7418
0.3461 4.4822 2900 0.5148 0.8444 0.7407
0.3228 4.6368 3000 0.5109 0.8338 0.7382
0.3222 4.7913 3100 0.5131 0.8473 0.7404
0.2552 4.9459 3200 0.5233 0.8435 0.7420
0.2804 5.1005 3300 0.5154 0.8435 0.7377
0.2257 5.2550 3400 0.5353 0.8357 0.7421
0.2731 5.4096 3500 0.5389 0.8512 0.7527
0.2638 5.5641 3600 0.5288 0.8512 0.7615
0.2674 5.7187 3700 0.5350 0.8483 0.7555
0.2701 5.8733 3800 0.5597 0.8454 0.7511
0.2733 6.0278 3900 0.5444 0.8367 0.7452
0.2106 6.1824 4000 0.5750 0.8261 0.7330
0.2631 6.3369 4100 0.5724 0.8338 0.7343
0.2227 6.4915 4200 0.5836 0.8145 0.7170
0.241 6.6461 4300 0.5633 0.8464 0.7544
0.2374 6.8006 4400 0.5758 0.8367 0.7437
0.2583 6.9552 4500 0.5661 0.8464 0.7475
0.2057 7.1097 4600 0.5843 0.8473 0.7521
0.1991 7.2643 4700 0.5905 0.8454 0.7466
0.1997 7.4189 4800 0.5916 0.8377 0.7421
0.1897 7.5734 4900 0.5918 0.8386 0.7454
0.2295 7.7280 5000 0.5916 0.8329 0.7391
0.2219 7.8825 5100 0.6007 0.8464 0.7511
0.2162 8.0371 5200 0.6132 0.8512 0.7550
0.1922 8.1917 5300 0.6047 0.8406 0.7474
0.206 8.3462 5400 0.6075 0.8473 0.7521
0.188 8.5008 5500 0.6161 0.8454 0.7474
0.2061 8.6553 5600 0.6218 0.8473 0.7494
0.2027 8.8099 5700 0.6237 0.8357 0.7382
0.2109 8.9645 5800 0.6226 0.8406 0.7462
0.2151 9.1190 5900 0.6235 0.8406 0.7461
0.1532 9.2736 6000 0.6253 0.8435 0.7470
0.1746 9.4281 6100 0.6257 0.8406 0.7440
0.1917 9.5827 6200 0.6288 0.8377 0.7411
0.1925 9.7372 6300 0.6298 0.8357 0.7395
0.1886 9.8918 6400 0.6299 0.8386 0.7423

Framework versions

  • Transformers 4.49.0
  • Pytorch 2.5.1+cu124
  • Datasets 3.3.2
  • Tokenizers 0.21.0
Downloads last month
196
Safetensors
Model size
112M params
Tensor type
F32
·
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.

Model tree for HrantDinkFoundation/turkish-hs-4class-prediction

Finetuned
(20)
this model