hebert-finetuned-hebrew-metaphor

The model is fine-tuned to determine if a word in a sentence is used metaphorically or literally.

The model was trained for the following verbs: לחלום, לחתוך, לעוף, לפרק, להדליק, לכבס, לכופף, לרסק, לבשל, למחוק, לקפוץ, לקרוע, לקצור, לרקוד, לשבור, לשדוד, לשתות, לטחון, לתפור, לזרוע

This model is a fine-tuned version of avichr/heBERT on HebrewMetaphors dataset. It achieves the following results on the evaluation set:

  • Loss: 0.4682
  • Accuracy: 0.9510

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 2e-05
  • train_batch_size: 16
  • eval_batch_size: 16
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 15

Training results

Training Loss Epoch Step Validation Loss Accuracy
No log 1.0 389 0.1813 0.9379
0.2546 2.0 778 0.2309 0.9479
0.08 3.0 1167 0.3342 0.9492
0.0298 4.0 1556 0.4076 0.9460
0.0298 5.0 1945 0.3803 0.9485
0.0105 6.0 2334 0.3674 0.9454
0.0077 7.0 2723 0.5356 0.9410
0.0088 8.0 3112 0.4776 0.9422
0.0044 9.0 3501 0.4258 0.9504
0.0044 10.0 3890 0.4305 0.9523
0.001 11.0 4279 0.4357 0.9548
0.0031 12.0 4668 0.4770 0.9473
0.0015 13.0 5057 0.4604 0.9523
0.0015 14.0 5446 0.4670 0.9510
0.0022 15.0 5835 0.4682 0.9510

Framework versions

  • Transformers 4.30.2
  • Pytorch 2.0.1+cu118
  • Datasets 2.13.1
  • Tokenizers 0.13.3

About Us

Created by Doron Ben-chorin, Matan Ben-chorin, Tomer Tzipori, Guided by Dr. Oren Mishali. This is our project as part of computer engineering studies in the Faculty of Electrical Engineering combined with Computer Science at Technion, Israel Institute of Technology. For more cooperation, please contact email:

Doron Ben-chorin: doronbh7@gmail.com

Matan Ben-chorin: matan.bh1@gmail.com

Tomer Tzipori: TomerTzipori@gmail.com

Downloads last month
5,533
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.

Dataset used to train tdklab/hebert-finetuned-hebrew-metaphor