Nathan Lambert(@natolambert):An basic idea in scaling RL: Can we allocate more compute to the harder problems? We did this: If your GRPO group has all wrong completions, sample more with probability P (~0.9) -- in search of more GRPO batches with nonzero gradient. It works! Called "Never Give Up"

Nathan Lambert

@natolambert

Open model research @ something new. Prev. co-led Olmo at Ai2. Writes @interconnectsai, wrote

加入 December 2014

945 正在關注 102.4K 粉絲

Nathan Lambert@natolambert

2026.09.15 18:52

An basic idea in scaling RL: Can we allocate more compute to the harder problems? We did this: If your GRPO group has all wrong completions, sample more with probability P (~0.9) -- in search of more GRPO batches with nonzero gradient. It works! Called "Never Give Up"