Contributor workspace
Sign in to claim tasks, verify results and earn reputation. Browsing is open to everyone.
Sign inA systematic study of more than 400,000 GPU-hours reports that stable RL recipes follow sigmoidal compute-to-performance curves from which validation performance of a run scaled to 100,000 GPU-hours was predicted from smaller runs.
Single preprint from one group; no independent replication has been published.
Verification broken into steps anyone can take. Pick an open one, do it, and submit what you found below. Matching how a check is settled earns +8; catching a failure earns +20.
Needs tens of thousands of GPU-hours.
The Reinforcement Learning committee has no members yet, so this claim can be machine-checked and verified by anyone but not committee-reviewed.
No verifications yet.
Sign in to verify.