Search papers, labs, and topics across Lattice.
This paper introduces Hierarchical Continuous Diffusion Language Models (H-CDLMs) by jointly diffusing token representations at different semantic granularities, inspired by discrete DLMs and image diffusion. This approach leverages parallel diffusion across modalities, including token-level and clustered token embeddings, with customizable per-modality samplers and schedules. H-CoBit, an instantiation of this framework, achieves state-of-the-art generative perplexity on LM1B and OWT, and significantly improves accuracy on GSM8K, outperforming prior continuous diffusion and flow-based models.
Jointly diffusing token and cluster embeddings unlocks significant gains in continuous diffusion language models, achieving state-of-the-art perplexity and reasoning performance with minimal overhead.
Diffusion Language Models (DLMs) hold the promise of order-agnostic, parallel text generation. Recently, continuous diffusion and flow matching models have seen substantial gains, driven by carefully crafted token representations and diffusion/flow spaces. In this work, we introduce Hierarchical Continuous Diffusion Language Models (H-CDLMs), a simple framework that further improves continuous DLMs with minimal compute and parameter overhead. Drawing on the discrete DLM and continuous image diffusion literature on joint diffusion, we diffuse multiple modalities in parallel. These modalities represent tokens at different semantic granularities: in our instantiation, the tokens themselves and coarser clusters obtained by clustering pretrained token embeddings. We propose a general setup that allows per-modality samplers and schedules to enhance the interplay between modalities. Applied to CoBit, this yields H-CoBit, which delivers large empirical gains across benchmarks. At dataset entropy, H-CoBit improves MAUVE and reaches a generative perplexity (GenPPL) of 49.4 on LM1B and 50.4 on OWT, improving on the baseline by 24.2 and 20.7 points and surpassing even discrete DLMs of comparable size. On GSM8K, it reaches 27.4% accuracy, outperforming prior continuous diffusion and flow-based models. We further apply H-CDLM to the flow matching model FLM, obtaining consistent gains with H-FLM and demonstrating that the framework generalizes across continuous generative paradigms. Our code will be made publicly available at https://github.com/matol-16/HCDLM.git .
