Saved in:
Bibliographic Details
Main Authors: Kong, Injin, Lee, Hyoungjoon, Jo, Yohan
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.14368
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914565400297472
author Kong, Injin
Lee, Hyoungjoon
Jo, Yohan
author_facet Kong, Injin
Lee, Hyoungjoon
Jo, Yohan
contents Continuous diffusion language models lag behind autoregressive transformers, partly because diffusion is applied in spaces poorly suited to language denoising and token recovery. We propose DiHAL, a geometry-guided diffusion-transformer hybrid that asks where diffusion should enter a pretrained transformer. DiHAL scores layers with geometry-based proxies, selects a diffusion-friendly hidden-state interface, and replaces the lower transformer prefix with a diffusion bridge while retaining the upper layers and original LM head. By reconstructing the selected-layer hidden state rather than tokens, DiHAL avoids direct continuous-to-discrete recovery. Experiments on 8B-scale backbones show that the geometry score predicts effective shallow insertion layers under a fixed bridge-training protocol and that hidden-state recovery improves over continuous diffusion baselines in a diagnostic comparison matching the diffusion/recovery training budget. These results suggest that hidden-state geometry helps identify where diffusion-based replacement is feasible inside pretrained language models.
format Preprint
id arxiv_https___arxiv_org_abs_2605_14368
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Where Should Diffusion Enter a Language Model? Geometry-Guided Hidden-State Replacement
Kong, Injin
Lee, Hyoungjoon
Jo, Yohan
Computation and Language
Artificial Intelligence
Continuous diffusion language models lag behind autoregressive transformers, partly because diffusion is applied in spaces poorly suited to language denoising and token recovery. We propose DiHAL, a geometry-guided diffusion-transformer hybrid that asks where diffusion should enter a pretrained transformer. DiHAL scores layers with geometry-based proxies, selects a diffusion-friendly hidden-state interface, and replaces the lower transformer prefix with a diffusion bridge while retaining the upper layers and original LM head. By reconstructing the selected-layer hidden state rather than tokens, DiHAL avoids direct continuous-to-discrete recovery. Experiments on 8B-scale backbones show that the geometry score predicts effective shallow insertion layers under a fixed bridge-training protocol and that hidden-state recovery improves over continuous diffusion baselines in a diagnostic comparison matching the diffusion/recovery training budget. These results suggest that hidden-state geometry helps identify where diffusion-based replacement is feasible inside pretrained language models.
title Where Should Diffusion Enter a Language Model? Geometry-Guided Hidden-State Replacement
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2605.14368