ELF-REG: Scaling Continuous Diffusion Language Models to Reasoning Tasks

ELF-REG, a method that scales Embedded Language Flows (ELF) continuous diffusion language models to reasoning tasks, reached 55.96% pass@1 on GSM8K at 64 network function evaluations and 13.39% on MATH-500 and 22.56% on HumanEval at 128 NFE, according to the arXiv paper 2609.29102v1. ELF-REG adds representation alignment and entanglement (REPA+REG), in which a frozen autoregressive teacher supervises intermediate denoiser features and supplies a global representation jointly denoised with the response, lifting MATH-500 pass@1 from 10.55% for the ELF-L baseline to 13.39% with ELF-REG-L. Without few-step training, the same task-specific checkpoints reach 41.21% HumanEval pass@10 at 16 NFE via early-stop decoding of an intermediate clean prediction.

arXiv:2609.29102v1 Announce Type: new Abstract: Fully continuous diffusion language models dLMs denoise continuous representations without intermediate discretization, then decode all response tokens in parallel at the final step. Their performance on challenging reasoning tasks remains less established than that of autoregressive AR LLMs and masked dLMs. We scale Embedded Language Flows ELF to mathematical reasoning and code generation on GSM8K, MATH-500, HumanEval, and MBPP. We introduce 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. ELF-REG-L achieves 55.96% pass@1 on GSM8K at 64 network function evaluations NFE , and 13.39% on MATH-500 and 22.56% on HumanEval at 128 NFE. It outperforms the evaluated comparable-scale dLMs in pass@1 on GSM8K and code, and improves MATH-500 pass@1 from 10.55% for the ELF-L baseline to 13.39% with ELF-REG-L. Without few-step training, the same task-specific checkpoints support strong low-NFE performance through early-stop, which decodes an intermediate clean prediction without completing the denoising trajectory. At 16 NFE, ELF-REG-L reaches 41.21% HumanEval pass@10, outperforming recent continuous dLMs of comparable scale.