Paper With Code Shuts Down, Open-Source Community Calls for a New All-in-One Resource Hub

With the official announcement of "Trending Papers" by Hugging Face co-founder and CTO Julien Chaumond on his X account,The news of Paper With Code closing down has been finalized.This "shocked" many developers and researchers and left them heartbroken.

The end of an era

From dataset titles and descriptions turning into garbled text to the site throwing “Bad Gateway 502” errors, Paper With Code had experienced outages and access issues since early July. Yet the team remained silent. Even user inquiries posted via GitHub Issues to its operations team went unanswered.

Long regarded as a resource hub for researchers,Paper With Code brought together papers, code, benchmarks, and leaderboards in one place,attracting a devoted global following. As more developers and researchers found themselves unable to access the site, speculation began to swirl across X, Reddit, LinkedIn, and other platforms — ranging from “Paper With Code has shut down” to “the site’s been attacked.”

It wasn’t until July 25 that Julien Chaumond, via his personal account, announced that Hugging Face had teamed up with Meta and Paper With Code to launch a replacement platform, Trending Papers— explicitly stating that Paper With Code was officially sunsetted the previous day.

The news quickly spread worldwide.Meta researcher Lucas Beyer put it bluntly: “End of an era.”

His post drew an outpouring of veteran users, who praised Paper With Code's SOTA and leaderboards features,and even described the site as “the Wikipedia of machine learning research.”

Similarly, there was also a lot of discussion about this on LinkedIn, with one PhD student commenting that: "Paper With Code (PWC) made it easier to discover great papers with runnable code that made those ideas real."

Can Hugging Face successfully take over?

The advantage of Paper With Code lies in its ability to integrate and organize resources.This categorized ranking, which breaks down models by field and presents them based on benchmark scores,helps users quickly identify the models they want to explore or benchmark against.

As a result, many users are not impressed by Hugging Face's heavily promoted Trending Papers feature.According to the main operators of Trending Papers, the section primarily focuses on tracking popular papers and linking them to their code implementations on GitHub. At present, it does not appear to have the ranking display feature that Paper With Code offers.“The current functionality is not detailed enough. I typically use Paper With Code's SOTA to identify the strongest models in different fields, which allows me to quickly determine which models our project should be compared against,” explained one user.

Some have also raised questions:"The 'paperwithcode' project has recorded information and comparisons for over 3k~5k different tasks. However, Hugging Face has only 40 tasks in the past few years. Is HG ready for this?"Why hasn't all this valuable information been replicated…"

Indeed, in the open-source domain, the importance and value of reproducible code are no less significant than theoretical innovations and breakthroughs.

Paper With Code has lived up to its name by providing reliable and neutral service in recent years,maintaining stable and neutral operations even after being acquired by Meta.The exact reasons behind the decision to sunset the platform remain unclear. It may be a reluctant move driven by questions about its commercial value, or perhaps the increasing operational pressure caused by the high frequency of updates to large models have become unsustainable...

However, it is certain that over its seven years of operation, Paper With Code has established a resource integration model highly recognized by the open-source community.Many developers have already voluntarily created alternative projects aiming to carry forward its legacy, but the challenges are evident, such as the massive scale and highly dispersed nature of resource aggregation and storage; efficient content updates amid the rapid evolution of machine learning and large models, and so on.While awaiting the refinement of Trending Papers' functionality, we also anticipate the emergence of more comprehensive resource platforms to rebuild the developers' resource hub.

About Paper With Code

Papers with Code was created in July 2018 with the original intention of helping machine learning enthusiasts track the latest published papers and source code, and quickly understand the most cutting-edge technological advances. The website matches the latest machine learning papers on arXiv with the code on GitHub (TensorFlow/PyTorch/MXNet, etc.). Users can search by title keywords or sort by popularity.

Founders Robert Stojnic and Ross Taylor both graduated from Cambridge University. Just one year after they co-founded the website, they were recruited by Meat AI (then Facebook AI) and jointly led the development of Llama 2 and Llama 3 at Meta AI.