Open Models Almanac
Methodspost-trainingagentic post-training

single mention · post-training

Weighted environment sampling

Weighted environment sampling assigns each RL training environment a sampling weight that governs how often the rollout threads draw its samples.1 This weights a mix of environments differently during training, letting some be deliberately upweighted relative to others.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. Each environment carries a sampling weight, the rate at which the rollout threads draw samples — tech report
  2. We deliberately upweight several agentic environments — tech report

Also called Weighted mix of RL training environments, Environment Mix.

Filed under post-training › agentic post-training