Fzz1/tb-training-mix-agentpick download history
Fzz1/tb-training-mix-agentpick is a reinforcement learning dataset on the Hugging Face Hub. In the last 30 days it was downloaded 700 times (509 in the last 7 days), and 700 times in total. It ranks #25,350 among datasets by monthly downloads.
tb-training-mix-agentpick Selected by an agent re-rank, not by text similarity. For each of 89 Terminal-Bench 2.1 tasks, an agent ranked ten candidate training tasks drawn from 5 corpora together — TMAX, RST, SWE-Smith, SWE-Rebench and TerminalWorld — scoring every candidate on skill,