Work

DGend/justgrpo-smaller-essential: v1.1.0

Year: 2026

Type: software

Abstract: Code accompanying the manuscript "Standard Group Relative Reinforcement Learning Is Unstable on Small Masked-Diffusion Language Models: A Controlled Study and Five Rejected Explanations," submitted to Scientific Reports. Includes the reinforcement learning driver, the supervised fine-tuning and evaluation scripts written for this work, and the deterministic analysis and figure-generation scripts that compute every statistic, figure, and table reported in the paper, together with unit tests for the analysis helpers. Model checkpoints and per-run training logs are not included; see the paper's Data Availability Statement.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.22104950

Authors: Kang, Young-Jin, Kim, Ki-Hwan, Jeong, Seok Chan

Institution: Dong-Eui University

Language: English

Topic: Topic Modeling

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7204271834

Cites:

Cited by:

Related to: 10