Hi guys,
I have fewer than 100 papers on my drive, and I think I’ve added annotations to around 50 of them. I believe this number is quite normal for researchers.
I subscribed to Paperpile on an annual plan, but the app on my laptop and iPad is too laggy to work with. This is very disappointing. Why charge users if you can’t even meet the basic requirements?
Welcome to our forum, @Bao_Dinh! Thanks for reaching out here as well as via email — as I mentioned there, we need more details in order to investigate. Please feel free to reply by either via so we can troubleshoot and get your library back up and running.
Hi Vicente,
I believe I have already described the details of my case. I don’t think I have a lot of annotated papers, but every time I try to open a PDF in Paperpile, it becomes extremely laggy and fails to load—especially for annotated ones. This happens even though I am using Chrome with only one tab open.
I am using a MacBook Air M2 with 8GB RAM. When I checked the Activity Monitor, I noticed a sudden spike to 10GB RAM usage, which then gradually drops.
@Bao_Dinh, I posed a few questions via mail:
The team confirms there are no glaring errors on backend, so it would be helpful if you could select one of the references that fails to load, press Ctrl+J and share the resulting output with us here or via email for further investigation. Let us know.
Hi Vicente,
Here are my answers to your questions:
• I think the slowness happens because too many documents and annotations have been added.
• There are no error messages, only lag.
• I have tried all configs of signing in and out, but it did not work.
Here are the messages from pressing Cmd + J on one of the references that fails to load:
[{
"_id": "54a3f00e-3744-04a9-a874-3a9869405509",
"abstract": "Recent advances in code-specific large language models (LLMs) have greatly enhanced code generation and refinement capabilities. However, the safety of code LLMs remains under-explored, posing potential risks as insecure code generated by these models may introduce vulnerabilities into real-world systems. Previous work proposes to collect security-focused instruction-tuning dataset from real-world vulnerabilities. It is constrained by the data sparsity of vulnerable code, and has limited applicability in the iterative post-training workflows of modern LLMs. In this paper, we propose ProSec, a novel proactive security alignment approach designed to align code LLMs with secure coding practices. ProSec systematically exposes the vulnerabilities in a code LLM by synthesizing error-inducing coding scenarios from Common Weakness Enumerations (CWEs), and generates fixes to vulnerable code snippets, allowing the model to learn secure practices through advanced preference learning objectives. The scenarios synthesized by ProSec triggers 25 times more vulnerable code than a normal instruction-tuning dataset, resulting in a security-focused alignment dataset 7 times larger than the previous work. Experiments show that models trained with ProSec is 29.2% to 35.5% more secure compared to previous work, with a marginal negative effect of less than 2 percentage points on model's utility.",
"archivePrefix": "arXiv",
"arxivid": "2411.12882",
"attachments": [
{
"_id": "932c9c79-3574-405d-a7fd-65511f8c400c",
"mimeType": "application/pdf",
"pub_id": "54a3f00e-3744-04a9-a874-3a9869405509",
"source_filename": "2411.12882v1.pdf",
"filename": "All Papers/X/Xu et al. 2024 - ProSec - Fortifying code LLMs with proactive security alignment.pdf",
"md5": "4bafb42db9240d738b80861cf2034ea2",
"article_pdf": 1,
"filesize": 803174,
"created": 1732840157.6969998,
"s3_md5": "4bafb42db9240d738b80861cf2034ea2"
}
],
"author": [
{
"last": "Xu",
"first": "Xiangzhe",
"initials": "X",
"formatted": "Xu X"
},
{
"last": "Su",
"first": "Zian",
"initials": "Z",
"formatted": "Su Z"
},
{
"last": "Guo",
"first": "Jinyao",
"initials": "J",
"formatted": "Guo J"
},
{
"last": "Zhang",
"first": "Kaiyuan",
"initials": "K",
"formatted": "Zhang K"
},
{
"last": "Wang",
"first": "Zhenting",
"initials": "Z",
"formatted": "Wang Z"
},
{
"last": "Zhang",
"first": "Xiangyu",
"initials": "X",
"formatted": "Zhang X"
}
],
"autoCleaned": 1,
"citekey": "Xu2024-rj",
"copyright": "http://creativecommons.org/licenses/by/4.0/",
"created": 1732840155.3439999,
"dup_sha1": "cad881defcd3013b5d2ab40dce3be2d43f86ce00",
"eprint": "2411.12882",
"folders": [
"1277aa6e-0b08-40bb-b135-fee963e4d01b"
],
"foldersNamed": [
"AI4Code/Secure Code Generation"
],
"id_list": [
"sha1:eaee59142a4a81c9f95c13b478f12496863c2acd",
"dup_sha1:cad881defcd3013b5d2ab40dce3be2d43f86ce00",
"arxivid:2411.12882",
"url:http://arxiv.org/abs/2411.12882"
],
"incomplete": 0,
"journal": "arXiv [cs.CR]",
"labels": [
"e4cf407f-30fc-4155-bf46-10d2992394c8",
"69a1d565-6ed4-44ec-acde-994305c95fd6"
],
"labelsNamed": [
"secure code generation",
"reinforcement learning"
],
"owner": "AFB924642C5011EFAE72458F945E4932",
"pdf_restricted": 0,
"primaryClass": "cs.CR",
"published": {
"year": "2024",
"month": "11",
"day": "19"
},
"pubtype": "PP_PREPRINT",
"sha1": "eaee59142a4a81c9f95c13b478f12496863c2acd",
"subfolders": [
"All Papers/X"
],
"title": "ProSec: Fortifying code LLMs with proactive security alignment",
"url": [
"http://arxiv.org/abs/2411.12882"
],
"__render": {
"title": [
{
"type": "<div>",
"className": "pp-grid-title",
"children": [
{
"type": "<span>",
"content": "ProSec: Fortifying code LLMs with proactive security alignment",
"className": "pp-grid-titletext"
}
]
}
],
"author": [
{
"type": "<span>",
"content": "Xu X",
"className": "author",
"data": "Xu X:~XIANGZHE~XU~"
},
", ",
{
"type": "<span>",
"content": "Su Z",
"className": "author",
"data": "Su Z:~ZIAN~SU~"
},
", ",
{
"type": "<span>",
"content": "Guo J",
"className": "author",
"data": "Guo J:~JINYAO~GUO~"
},
", ",
{
"type": "<span>",
"content": "Zhang K",
"className": "author",
"data": "Zhang K:~KAIYUAN~ZHANG~"
},
", ",
{
"type": "<span>",
"content": "Wang Z",
"className": "author",
"data": "Wang Z:~ZHENTING~WANG~"
},
", ",
{
"type": "<span>",
"content": "Zhang X",
"className": "author",
"data": "Zhang X:~XIANGYU~ZHANG~"
}
],
"journal": "arXiv [csCR]",
"pubtype": "Preprint"
}
}]
Btw, many long documents fail to load, not just some.
Thanks for the details, @Bao_Dinh. It seems the issue might lie in the amount of annotations, but since this has not happened before we need the actual PDF in order to reproduce. Would it be possible for you to share with us? We won’t share or keep your files beyond troubleshooting.
To do this, select one or a couple of the affected refs in your library, press Shift+S and then copy/paste the link here. Make sure the options Share PDF and other files and Share notes and annotations are ticked like below.
Let us know, and thanks again for your patience and help.
