bert-paper-classifier-arxiv – AI Model by voroninip | AlphaNeural AI

You can deploy this model and start earning money today!

voroninip/
bert-paper-classifier-arxiv

like

0

text-classification

generated_from_trainer

arxiv-community/arxiv_dataset

microsoft/deberta-v3-base

autotrain_compatible

endpoints_compatible

Views

No views yet

bert-paper-classifier-arxiv

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

Loss: 1.1651
Accuracy: 0.6854

Model description

Predict category of the paper based on its title and article.

Training and evaluation data

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 128
eval_batch_size: 16
seed: 42
gradient_accumulation_steps: 3
total_train_batch_size: 384
optimizer: Use OptimizerNames.ADAFACTOR and the args are: No additional optimizer arguments
lr_scheduler_type: cosine
num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	97	1.5356	0.6095
2.3002	2.0	194	1.3083	0.6656
1.5053	3.0	291	1.2135	0.6822
1.3177	4.0	388	1.1702	0.6846
1.1984	4.9550	480	1.1651	0.6854

Framework versions

Transformers 4.50.3
Pytorch 2.6.0+cu124
Datasets 3.5.0
Tokenizers 0.21.1