Latent-DARM: Bridging Discrete Diffusion And Autoregressive Models For Reasoning
This work advances multi-agent collaboration for reasoning tasks by enabling heterogeneous models to work together more effectively, though it is incremental in combining existing model types.
The paper tackles the problem of limited global reasoning in autoregressive language models (ARMs) and poor text fluency in discrete diffusion language models (DDLMs) by introducing Latent-DARM, a latent-space communication framework that bridges DDLM planners and ARM executors for multi-agent collaboration, resulting in improved accuracy from 27.0% to 36.0% on DART-5 and from 0.0% to 14.0% on AIME2024 while using less than 2.2% of the token budget of state-of-the-art models.
Most multi-agent systems rely exclusively on autoregressive language models (ARMs) that are based on sequential generation. Although effective for fluent text, ARMs limit global reasoning and plan revision. On the other hand, Discrete Diffusion Language Models (DDLMs) enable non-sequential, globally revisable generation and have shown strong planning capabilities, but their limited text fluency hinders direct collaboration with ARMs. We introduce Latent-DARM, a latent-space communication framework bridging DDLM (planners) and ARM (executors), maximizing collaborative benefits. Across mathematical, scientific, and commonsense reasoning benchmarks, Latent-DARM outperforms text-based interfaces on average, improving accuracy from 27.0% to 36.0% on DART-5 and from 0.0% to 14.0% on AIME2024. Latent-DARM approaches the results of state-of-the-art reasoning models while using less than 2.2% of its token budget. This work advances multi-agent collaboration among agents with heterogeneous models.