Mo

Results 10 comments of Mo

I will try to look into that..

Are these the ones shared on www.swebench.com or different?

"This repository contains the predictions, execution logs, trajectories, and results for model inference + evaluation runs on the [SWE-bench](https://swe-bench.github.io/) task." https://github.com/swe-bench/experiments

Thanks for the tip @joshkyh

Maybe a fixed web link of the benchmark image so that it updates everywhere whenever there is a new agent.

> multiple tries, Range?

> In our experiments, the best models can achieve rather high scores when given multiple tries, and I believe that the practical upper bound is much higher than the current...

> I do not think this is possible anymore. And that is a good thing indeed! (unless you are on openBSD or something of that sort)

@qcgm1978 looks amazing! can't wait to try it myself. Thank you for the hardwork.