abbassix/ComNumPNdistilBERTv1-big

ComNumPNdistilBERT

This model is a fine-tuned version of distilbert-base-uncased a subset (8000 samples as training and 2000 samples as validation) of the ComNum dataset. We changed every single digit on a certain position in the numeral into a special character. For example '26753' becomes '3𐄇 5𐄈 7𐄉 6𐄊 2𐄋'. Since the order of magnitude of numerals in the test set is 6, while that of numerals in training and evaluation sets are from 0 to 5, and the fine-tuned model has never seen special character for digits of the order of magnitude of 6 (e.g., '1𐄏', '6𐄏' or '9𐄏'), it cannot generalize well.

It achieves the following results on the evaluation set:

• Loss: 0.0458
• Accuracy: 0.9875

It achieves the following results on the test set:

• Loss: 3.0438
• Accuracy: 0.5909

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-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
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.2951	1.0	1000	0.1900	0.886
0.09	2.0	2000	0.0701	0.9835
0.0382	3.0	3000	0.0458	0.9875

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.0
• Tokenizers 0.15.0