---
format: "aidr-story-markdown/v1"
id: "160599d47f7014e3b3b4f56970e6b95de6eec38f9a36a10ce292827e2bd6951f"
canonical_url: "https://aidr.today/160599d4?lang=en"
title: "TraceQuery: learning what to recompute from full-denoising trajectory in diffusion language models"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-27T04:00:00.000Z"
category: "Research"
topics: ["inference"]
source_urls: ["https://arxiv.org/abs/2610.33008"]
summary: "Researchers propose TraceQuery, a method for efficient inference in diffusion language models that formulates dLLM inference as trajectory-guided querying. The approach leverages the full-denoising trajectory to determine which token representations to recompute, reducing inference cost while maintaining accuracy by keeping generation close to the full-denoising trajectory."
---

# TraceQuery: learning what to recompute from full\-denoising trajectory in diffusion language models

> [Open the canonical story](<https://aidr.today/160599d4?lang=en>)

**Published:** 2026-10-27T04:00:00.000Z
**Category:** Research
**Topics:** inference

## Summary

Researchers propose TraceQuery, a method for efficient inference in diffusion language models that formulates dLLM inference as trajectory\-guided querying\. The approach leverages the full\-denoising trajectory to determine which token representations to recompute, reducing inference cost while maintaining accuracy by keeping generation close to the full\-denoising trajectory\.

## Sources

- [Story source](<https://arxiv.org/abs/2610.33008>)