SDAR-4B

The 'SDAR-4B' dataset/benchmark contains a collection of tasks used to evaluate the performance of reinforcement learning methods for diffusion large language models, specifically in terms of their accuracy on mathematical reasoning benchmarks like MATH500 and GSM8K.

Papers using SDAR-4B (1)

SLIM-RL: Risk-Budgeted Random-Masking RL for Diffusion LLMs Without Trajectory Slicing2026