Prepare for an arXiv submission

Submission Steps

Download source code from Overleaf if you use it: menu -> download -> source.
Strip comments and combine all tex files (f01-main.tex, f02-intro.tex, etc.) into one file arxiv_main.tex.

# Replace f01-main.tex with the main tex file in your overleaf project
latexpand --empty-comments f01-main.tex > arxiv_main.tex

Create a new directory and copy only essential files into it (arxiv_main.tex, style files, figures, etc.).
Compile arxiv_main.tex with BibTeX in your new directory by running the following four commands. The main purpose of this step is to generate a .bbl file. You would need to run pdflatex three times. Check if the generated arxiv_main.pdf looks correct.

pdflatex arxiv_main
bibtex arxiv_main
pdflatex arxiv_main
pdflatex arxiv_main

Zip all files inside the directory (don't include the directory itself). Upload the tarball file arxiv.tar.gz to arXiv.

tar czvf arxiv.tar.gz ./*

Submissions received between (all times Eastern US)	Will be announced (all times Eastern US)	Mailed to subscribers
Monday 14:00 – Tuesday 14:00	Tuesday 20:00	Tuesday night / Wednesday morning
Tuesday 14:00 – Wednesday 14:00	Wednesday 20:00	Wednesday night / Thursday morning
Wednesday 14:00 – Thursday 14:00	Thursday 20:00	Thursday night / Friday morning
Thursday 14:00 – Friday 14:00	Sunday 20:00	Sunday night / Monday morning
Friday 14:00 – Monday 14:00	Monday 20:00	Monday night / Tuesday morning

If you post between Wednesday 14:00 – Thursday 14:00, it will be announced on Thursday 20:00 and be on the arXiv front-page for three days. This is because the next announcement will be on Sunday 20:00. This is the best day to post on arXiv if you want to maximize the exposure.
By default, arXiv orders submissions by "recentness". Therefore, if you submit right before 14:00, it is more likely to be listed on the first page of the list (top 25 most recent submissions). On Thursday, there are usually about 70-100 machine learning submissions (3-4 pages!)

⚠️ latexpand --empty-comments does not remove comments in \begin{comment}\end{comment}. If you write comments other than the native way %, you can use arxiv-latex-cleaner to clean comments.
Check out the checklist for arXiv submission from Overleaf (Thanks @bhoov)
Remember to keep your primary .tex file in the root level of the repository
natbib in an unsorted number style does not work on ArXiv. Solution: remove natbib as a dependency, use \bibliographystyle{unsrt} instead of \bibliographystyle{unsrtnat}

Now arXiv can successfully compile your LeTeX file! 🎉 Here are some tips to make the generated PDF look great.

Double check the generated PDF. It's likely that arXiv is using a different LaTeX version from yours, resulting in a different-looking PDF.
1. If you use PDFLaTeX to compile LaTeX files locally, you can set \pdfoutput=1 in the start of your arxiv_main.tex to force arXiv to also use PDFLaTeX.
2. Sometimes arXiv would mess up with the microtype for full justification papers, especially in section headers and figure captions (e.g., adding a new line). Double check the generated PDF and modify the text when needed. One way to avoid new lines is to wrap your text in \mbox{} and force LaTeX to overflow that line.
Add page numbers. Typically conference style files hide page numbers, so that all accepted papers can bundled into a conference proceeding with a proceeding page number. However, preprints with page numbers are easier to read. Here is a simple guide to add page number for different popular CS conference LaTeX templates.
Use the correct copyright block (especially for ACM conferences). Make sure your preprint PDF claims the correct copyright license (e.g., CC, IW3C2W3, rightsretained). For ACM conferences, you can change the copyright block with the command setcopyright{}. However, if your paper has a Creative Common (CC) license, you would need some hacks to make it work.