Open Models Almanac
Methodspost-trainingreward modelling

single mention · post-training

Multi-level reward formulation

A multi-level reward formulation partitions reward signals in the HTML-based slide generation process into three levels.1 It evaluates properties ranging from static markup attributes, such as color and spacing, to runtime rendering metrics like element width and height.2,3

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 propose a multi-level reward formulation, which partitions reward signals in the HTML-based slide generation process into three levels — GLM-5: from Vibe Coding to Agentic Engineering
  2. This level focuses on declarative attributes in the generated HTML, including positioning, spacing, color, typography, saturation, and other stylistic attributes — GLM-5: from Vibe Coding to Agentic Engineering
  3. Unlike static inspection, this level evaluates runtime properties of DOM nodes during rendering, such as element width and height, bounding boxes, and other geometric layout metrics — GLM-5: from Vibe Coding to Agentic Engineering

Filed under post-training › reward modelling