Kevin Li releases largest open agentic trace dataset
Kevin Li, an MS CS student at Stanford, released SWE-ZERO-12M-trajectories, the largest open agentic trace dataset. The collection holds 12 million trajectories and 112 billion tokens drawn from 122,000 pull requests across 3,000 repositories in 16 programming languages. It is 5.7 times larger than the previous leading open dataset and is hosted on Hugging Face. Researchers including Percy Liang and David Hall reposted the announcement.
#25Percy Liang@PERCYLIANGGoing into the next Marin run.
Kevin Li@kevin_x_liIntroducing SWE-ZERO-12M-trajectories: the largest agentic trace dataset in the open, 5.7x larger than the previous largest. 112B tokens · 12M trajectories · 122K PRs · 3K repos · 16 languages https://huggingface.co/datasets/AlienKevin/SWE-ZERO-12M-trajectories
#936John Yang@JYANGBALLINSWE-ZERO uses @KLieret and I's mini-SWE-agent as the agent harness
Absolutely mindblowing the scale at which SWE-bench data has scaled.
A year ago, SWE-smith's 50k instances across 128 repos. @kevin_x_li's SWE-Zero blows this out the water. 👏
Kevin Li@kevin_x_liIntroducing SWE-ZERO-12M-trajectories: the largest agentic trace dataset in the open, 5.7x larger than the previous largest. 112B tokens · 12M trajectories · 122K PRs · 3K repos · 16 languages https://huggingface.co/datasets/AlienKevin/SWE-ZERO-12M-trajectories