ASSERT-KTH / VRepair

open science repo of "Neural Transfer Learning for Repairing Security Vulnerabilities in C Code" https://arxiv.org/pdf/2104.08308
57 stars 18 forks source link

Hardwares used for experiments #6

Closed dangnguyenngochai closed 9 months ago

dangnguyenngochai commented 2 years ago

Can you share the hardware specifications used for experiments, along with the time it takes for transfer learning in rq2 and pre-training + target domain training in rq3 ?

chenzimin commented 2 years ago

We used NVIDIA Tesla T4 to train all the models.

transfer learning in rq2

source domain training (training on a bug fix corpus) takes ~3 - 7 days, target domain training (truing on a vulnerability fix corpus) takes ~1h.

pre-training + target domain training in rq3

pre-training take ~3-7 days, and target domain training takes similar time as before.

dangnguyenngochai commented 2 years ago

How about your other method, SeqTrans, how long does it take for training using the specifications mentioned in your paper "Intel Xeon E5 processor, four Nvidia 3090 GPU, and 1TB RAM"

monperrus commented 2 years ago

How about your other method, SeqTrans

We're not authors of SeqTrans.

dangnguyenngochai commented 2 years ago

Very sorry, my mistake.

dangnguyenngochai commented 2 years ago

Can you share the trained models for your research questions? It would be nice if I don't have to wait that long to test some of my ideas.

chenzimin commented 2 years ago

I am now contacting Zenodo to host all the models (compressed version is 273GB).

I have them currently uploaded to OneDrive for now:

I will update the repo about the model directory structure.

dangnguyenngochai commented 2 years ago

How long can you let these models be accessible via one drive? I don't think I can transfer the whole thing to my server for now.

chenzimin commented 2 years ago

The link will be valid until Sep 2022. The models can be stored at OneDrive at least until Oct 2023, in the meantime I will try to find a location to host the models.

kakairesteven commented 11 months ago

Greetingas @chenzimin, much thanks for the research work. I wish to have a downloadable copy of the model. Thanks.

monperrus commented 9 months ago

The models have been pushed to Zenodo, see README.