ELF-REG: Scaling Continuous Diffusion Language Models to Reasoning Tasks
This work introduces ELF-REG, which improves learning with representation alignment and entanglement (REPA+REG), where a frozen AR teacher supervises intermediate denoiser features and supplies a global representation that is jointly denoised with the response.
Z. Li, William Chen, Bing-Shuo Qian et al.
· 0 citations