paperswithcode
https://paperswithcode.com now redirects to huggingface.
Have Meta and Huggingface seriously killed paperswithcode?
I don't consider the Trending Papers page a proper replacement. This is most likely causing serious harm to the community in general.
🚀 We’re excited to announce the beta launch of SOTAPapers.com
SOTAPapers is a modern, improved alternative to Papers With Code — a platform for exploring and discovering state-of-the-art research in AI and machine learning. Like Papers With Code, it organizes and connects research papers, datasets, methods, tasks, benchmarks, and venues — but with a cleaner structure, faster search, and better filtering to help you quickly evaluate progress across domains.
Whether you’re a researcher, engineer, or academic, SOTAPapers offers a structured view of the latest advancements, making it easier than ever to stay on top of the field.
Importing the published Papers With Code database was no small task — the data was inconsistently structured and took significant effort to clean, organize, and improve. But now it’s live, and we need your help to fill in the gaps, hunt bugs, and make it even better.
You can contribute data submissions or report bugs via our GitHub: [GitHub link here] We’re looking forward to your feedback and ideas!
For anyone else looking for an answer and explanation, here is the announcement and screenshot of it. Hugging Face and Meta have shut down Papers with Code, and the link now redirects to the Trending page on Hugging Face. The Trending tab has no categories, no benchmarks or historical performance of models, and no real way to discover research beyond whatever is manually highlighted.
But, good news! Dino v3 from Meta is the number one paper, as it has been for the 5 days since its release!
https://x.com/julien_c/status/1948718297194561956
@ma-abdellaoui i am looking at the website ,currently the search needs to be refined,for instance the dataset RPLAN doesnt show up on your list.
HuggingFace Daily Paper does not have any features of PapersWithCode. They just eat up. Another day, another monopoly.
I don't care about Hugging Face, but I can't accept the absence of Papers With Code. Why would such a good site be shut down?
Such a sad day for science. I hope arxiv or semanticscholar takes this chance to introduce a public-funded initiative, instead of these corporate leeches pretending to do public good.
Hey friends, we did not pretend we had built a 1:1 replacement to Papers With Code (which we loved, same as you). Our Trending papers section is only a partial alternative to PWC but one that some community members have found useful, so we chimed in and proposed Meta to redirect to it (instead of just killing the domain name)
We plan to build some other PWC-inspired features over time but bear with us in the meantime – and let's try to keep this discussion constructive!
Hey friends, we did not pretend we had built a 1:1 replacement to Papers With Code (which we loved, same as you). Our Trending papers section is only a partial alternative to PWC but one that some community members have found useful, so we chimed in and proposed Meta to redirect to it (instead of just killing the domain name)
We plan to build some other PWC-inspired features over time but bear with us in the meantime – and let's try to keep this discussion constructive!
@julien-c what is the best way to request new features from the new PWC hosting by huggingface? In particular I'm missing a web feed (requested also at https://github.com/ma-abdellaoui/sotapapers-contribute/issues/5)
We are a group of AI researchers who are currently in school. We have almost 1:1 replicated the functions of paper with code, and it almost includes all the papers. Some pictures are missing, but most of the data and code, as well as the rankings, are still there. I hope to address everyone's temporary needs. http://scholarwiki.ai/
Can someone give me the lowdown on the “mysterious alliance” between Papers with Code, Hugging Face, and Meta?
I’ve always found Papers with Code super useful—it’s basically my only compass for testing open-source models. But today I discovered it’s completely vanished from search engines, and the links magically redirect to Hugging Face… what sorcery is this?
If this is a Hugging Face move, I swear I’m never renewing Hugging Face Pro again! 🤬
I found this explanation on X from rosstaylor90: https://x.com/rosstaylor90/status/1948728240526557580
Is this still working? It's not linking
We are a group of AI researchers who are currently in school. We have almost 1:1 replicated the functions of paper with code, and it almost includes all the papers. Some pictures are missing, but most of the data and code, as well as the rankings, are still there. I hope to address everyone's temporary needs. http://scholarwiki.ai/
I found this explanation on X from rosstaylor90: https://x.com/rosstaylor90/status/1948728240526557580
that's not a proper explanation, it's still as mysterious as before ... my take: meta wants to slow down public progress and maybe even uses the platform internally instead
Bro this is insane, they (HuggingFace) clam to me for a community and people, free access and everything. But all I see it's a walled garden, that consumes and destroys all that actually useful for community. Their site is so convoluted to use, even that I spent my whole life in front of a PC. Amount of redirects, registrations, and gatekeeping on that site make me vоmit.
I bet they absored papers with code into their "Enterprise/Pricing" buttons that constantly on top of the site.
Hey friends, we did not pretend we had built a 1:1 replacement to Papers With Code (which we loved, same as you). Our Trending papers section is only a partial alternative to PWC but one that some community members have found useful, so we chimed in and proposed Meta to redirect to it (instead of just killing the domain name)
We plan to build some other PWC-inspired features over time but bear with us in the meantime – and let's try to keep this discussion constructive!
If I could only make one request, please try to surface as much of the original data from papers with code rather than just dropping those items from your new DB. There are many links that point to the original papers with code (including from the tensorflow docs for a lot of datasets and other resources). These are now effectively dead links since the papers referenced aren't even in the new system. If you're not going to re-host those same pages, even a simple redirect to the DOI link of the original paper would probably be a better experience.
I'll accept that I could be wrong and might have not seen them in the search results, but it's hard to say when there's only a single box saying "search with AI" and no way to filter by category, paper type, or other criteria. Those sorts of features would make it much easier to navigate if you do come to the site from a papers with code link.
There are many links that point to the original papers with code (including from the tensorflow docs for a lot of datasets and other resources). These are now effectively dead links since the papers referenced aren't even in the new system. If you're not going to re-host those same pages, even a simple redirect to the DOI link of the original paper would probably be a better experience.
As a friday night project, I've set up redirects for the paper pages URLs from Paperswithcode to the corresponding ArXiv-indexed page on HF
Example: https://paperswithcode.com/paper/cogagent-a-visual-language-model-for-gui now redirects to https://huggingface.co/papers/2312.08914
Hopefully should already reduce a bit the number of broken links in the wild!
As a friday night project, I've set up redirects for the paper pages URLs from Paperswithcode to the corresponding ArXiv-indexed page on HF
Example: https://paperswithcode.com/paper/cogagent-a-visual-language-model-for-gui now redirects to https://huggingface.co/papers/2312.08914
Hopefully should already reduce a bit the number of broken links in the wild!
Thanks for putting this together so quickly for both papers and datasets. This is a huge help, and it looks to be working well for the links I've tried from the TensorFlow docs.
I hope that the search experience for hugging face's papers repository continues to expand in its power and capabilities, given that it's still relatively recent, but this definitely helps a bit in the meantime, since the original paperswithcode links still show up often in search engine results.