Open Models Almanac
Methodspost-trainingreinforcement learning algorithm

specific method · filed under post-training

Groupwise Advantage Redistribution

A reinforcement learning method that redistributes sequence-level advantages within a rollout group using quality rankings of the trajectories.1 An agentic grader jointly examines all trajectories in the group, contrasting successes and failures and ranking passing ones along quality dimensions, and these assessments guide the ranking.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. We use the remaining quality rankings to redistribute sequence-level advantages — MiMo V2 6 technical report
  2. An SFT-trained agentic grader jointly examines all trajectories within the group, contrasting successful and failed attempts — MiMo V2 6 technical report

Also called Groupwise Advantage Redistribution (GAR).

sources
3
models
2
lab adopt it
1
strongest
core

Documented in

Evidence

4 quotes from 3 sources, one card per model, each opening with its strongest.

How sources treat it
used 3core 1

MiMo-V2.6-Flash

Xiaomiadopted2 sources

Groupwise Advantage Redistribution (GAR) ranks passing trajectories online and moves advantage toward higher-quality solutions.

coreoptimizationin MiMo-V2.6-Flash-RLXiaomi
2 more from this model

Through our proposed Groupwise Reward Synthesis (GRS) and Groupwise Advantage Redistribution (GAR), these fine-grained distinctions are converted into more informative learning signals

usedtraining objectivein MiMo-V2.6 RL trainingXiaomi

Its online agentic grader jointly examines successful and failed trajectories within each mixed-outcome group, ranks passing solutions, and redistributes positive advantage toward higher-quality passing trajectories.

usedpost trainingin MiMo-V2.6-ProXiaomi

MiMo-V2.6-Pro

Xiaomiadopted2 sources

Through our proposed Groupwise Reward Synthesis (GRS) and Groupwise Advantage Redistribution (GAR), these fine-grained distinctions are converted into more informative learning signals

usedtraining objectivein MiMo-V2.6 RL trainingXiaomi
2 more from this model

Its online agentic grader jointly examines successful and failed trajectories within each mixed-outcome group, ranks passing solutions, and redistributes positive advantage toward higher-quality passing trajectories.

usedpost trainingin MiMo-V2.6-ProXiaomi

Groupwise Advantage Redistribution (GAR) ranks passing trajectories online and moves advantage toward higher-quality solutions.

usedoptimizationin MiMo-V2.6-Pro-RLXiaomi
↑