Aggressive RL config sparks debate: 25k rollouts per step

A discussion erupted over an aggressive RL training config of 30 steps with 25,000 rollouts per step at async-4, called 'WILD' by Andrew Carr, with respondents arguing that the number of steps, not rollouts, is the key factor.

2026-09-22 ~ 2026-09-22 · 3 related posts

• A Wild Async RL Config: 30 Steps x 25K Rollouts Per Step at Parallelism 4 — willcb · 2026-09-22
• RL config with 25k rollouts per step over just 30 steps called 'WILD' — andrew_n_carr · 2026-09-22
• RL training config debate: 30 steps x 25k rollouts is wild, steps ≈ rollouts is the sane default — willcb · 2026-09-22