Model techniques map
Techniquespost-trainingreward modelling

general family · filed under post-training

Generative Reward Model

A reward model that generates evaluations rather than returning only a scalar reward; the evidence also describes rubric-based response and trajectory scoring.

Also called GenRM, Generative Reward Model (GRM), Rubric-conditioned Generative Reward Model.

sources
6
models
4
labs adopt it
2
strongest
used

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 7

Documented in

Further reading

Picked by hand, not extracted: where to read more, not evidence for anything on this page.

Evidence

7 spans quoted from the sources, strongest treatment first.

GenRM used for RLHF

usedpost trainingin Nemotron 3 UltraNVIDIA

We then develop detailed evaluation rubrics across multiple quality dimensions and employ a generative reward model to score responses based on these rubrics.

useddata curationin DeepSeek-V3.2DeepSeek

For general tasks, we employ a generative reward model where each prompt has its own rubrics for evaluation.

usedtraining objectivein DeepSeek-V3.2DeepSeek

we curate rubric-guided RL data and employ a Generative Reward Model (GRM) to evaluate policy trajectories.

usedpost trainingin DeepSeek-V4DeepSeek

Nemotron 3 Ultra 550B-A55B GenRM: GenRM used for RLHF

usedpost trainingin Nemotron 3 UltraNVIDIA

Generative Reward Model replaces scalar reward models

usedpost trainingin DeepSeek-V4DeepSeek

Generative Reward Models (GenRM) (Wang2025HelpSteer3Preference) with reasoning capabilities help mitigate such reward hacking behaviors

usedoptimizationin Nemotron 3 Ultra (Chat teacher)NVIDIA

Filed alongside

Other methods under post-training :: reward modelling.