Revisiting Judge Decoding from First Principles via Training-Free Distributional Divergence
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911360699334656 |
|---|---|
| author | Sun, Shengyin Li, Yiming Liu, Renxi Lin, Weizhe Zhen, Hui-Ling Yu, Xianzhi Yuan, Mingxuan Ma, Chen |
| author_facet | Sun, Shengyin Li, Yiming Liu, Renxi Lin, Weizhe Zhen, Hui-Ling Yu, Xianzhi Yuan, Mingxuan Ma, Chen |
| contents | Judge Decoding accelerates LLM inference by relaxing the strict verification of Speculative Decoding, yet it typically relies on expensive and noisy supervision. In this work, we revisit this paradigm from first principles, revealing that the ``criticality'' scores learned via costly supervision are intrinsically encoded in the draft-target distributional divergence. We theoretically prove a structural correspondence between learned linear judges and Kullback-Leibler (KL) divergence, demonstrating they rely on the same underlying logit primitives. Guided by this, we propose a simple, training-free verification mechanism based on KL divergence. Extensive experiments across reasoning and coding benchmarks show that our method matches or outperforms complex trained judges (e.g., AutoJudge), offering superior robustness to domain shifts and eliminating the supervision bottleneck entirely. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_04766 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Revisiting Judge Decoding from First Principles via Training-Free Distributional Divergence Sun, Shengyin Li, Yiming Liu, Renxi Lin, Weizhe Zhen, Hui-Ling Yu, Xianzhi Yuan, Mingxuan Ma, Chen Computation and Language Judge Decoding accelerates LLM inference by relaxing the strict verification of Speculative Decoding, yet it typically relies on expensive and noisy supervision. In this work, we revisit this paradigm from first principles, revealing that the ``criticality'' scores learned via costly supervision are intrinsically encoded in the draft-target distributional divergence. We theoretically prove a structural correspondence between learned linear judges and Kullback-Leibler (KL) divergence, demonstrating they rely on the same underlying logit primitives. Guided by this, we propose a simple, training-free verification mechanism based on KL divergence. Extensive experiments across reasoning and coding benchmarks show that our method matches or outperforms complex trained judges (e.g., AutoJudge), offering superior robustness to domain shifts and eliminating the supervision bottleneck entirely. |
| title | Revisiting Judge Decoding from First Principles via Training-Free Distributional Divergence |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2601.04766 |