Use asyncio instead of threadpoolexecutor https://github.com/OpenHands/benchmarks/pull/446 @millions-crayon-32880 @limited-student-10747 Validated on multiple runs. The (successful) big swtbench run was on a previous commit so I retriggered one that is on-going. swtbench runs https://github.com/OpenHands/benchmarks/pull/446#issuecomment-4091154611

I only see a big one from today https://github.com/OpenHands/benchmarks/actions/runs/23299736781 but it's mostly cached Total Images: 433 Successful: 433 ✅ Built: 0 🛠️ Skipped: 433 ⏭️

yes I used your previous run SHAs to skip image building. THis is for OOMing during evaluation, not image building!

Awesome!