DiHAL uses geometry proxies to pick where to replace the lower layers of a pretrained transformer with a diffusion bridge for hidden-state reconstruction, improving over token-level diffusion baselines on 8B models.
Title resolution pending
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
citation-role summary
baseline 1
citation-polarity summary
roles
baseline 1polarities
baseline 1representative citing papers
Spherical vMF flows reduce the continuity equation on the sphere to a scalar ODE in cosine similarity, enabling posterior-weighted sampling of categorical sequences via cross-entropy trained posteriors.
Absorbing discrete diffusion models the conditional distributions of clean data; reparameterizing yields a time-independent RADD that unifies with AO-ARMs and reaches SOTA perplexity among diffusion models on zero-shot language benchmarks.
Categorical flow matching models scale to 1.7B parameters on 2.1T tokens, enabling 4-step text generation with competitive quality and benchmark performance.
citing papers explorer
-
Where Should Diffusion Enter a Language Model? Geometry-Guided Hidden-State Replacement
DiHAL uses geometry proxies to pick where to replace the lower layers of a pretrained transformer with a diffusion bridge for hidden-state reconstruction, improving over token-level diffusion baselines on 8B models.
-
Spherical Flows for Sampling Categorical Data
Spherical vMF flows reduce the continuity equation on the sphere to a scalar ODE in cosine similarity, enabling posterior-weighted sampling of categorical sequences via cross-entropy trained posteriors.
-
Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data
Absorbing discrete diffusion models the conditional distributions of clean data; reparameterizing yields a time-independent RADD that unifies with AO-ARMs and reaches SOTA perplexity among diffusion models on zero-shot language benchmarks.
-
Scaling Categorical Flow Maps
Categorical flow matching models scale to 1.7B parameters on 2.1T tokens, enabling 4-step text generation with competitive quality and benchmark performance.