@inproceedings{luo-etal-2026-grove,
title = "{GROVE}: Graph-based Routing Over Variant Experts for Adaptive Two-Stage Training of Reasoning Models",
author = "Luo, Yiming and
Tian, Chunlin and
Li, Li",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1332/",
doi = "10.18653/2026.findings-emnlp.1332",
pages = "28210--28224",
ISBN = "979-8-89176-462-0",
abstract = "Post-training reasoning models is computationally expensive, and the allocation between supervised fine-tuning (SFT) and reinforcement learning (RL) substantially affects reasoning quality. Existing methods operate at the training-procedure level and treat the reasoning corpus as homogeneous, ignoring the structural heterogeneity of individual traces. We propose GROVE (Graph-based Routing Over Variant Experts), an adaptive framework that routes each example to SFT or RL based on the structural topology of its reasoning trace. GROVE parses each trace into a Reasoning Tree, scores its complexity via graph entropy, and partitions the corpus through a data-driven Gaussian Mixture Model with no hand-tuned threshold. Low-complexity traces are trained with SFT and high-complexity traces with GRPO as parallel LoRA adapters from a shared frozen base, merged into a single deployable model with no inference-time overhead. Across two training corpora and five reasoning benchmarks, GROVE consistently outperforms single-stage and naive-hybrid baselines, and a reversed-routing ablation confirms that this mapping is intrinsic rather than arbitrary."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="luo-etal-2026-grove">
<titleInfo>
<title>GROVE: Graph-based Routing Over Variant Experts for Adaptive Two-Stage Training of Reasoning Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yiming</namePart>
<namePart type="family">Luo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chunlin</namePart>
<namePart type="family">Tian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Li</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Post-training reasoning models is computationally expensive, and the allocation between supervised fine-tuning (SFT) and reinforcement learning (RL) substantially affects reasoning quality. Existing methods operate at the training-procedure level and treat the reasoning corpus as homogeneous, ignoring the structural heterogeneity of individual traces. We propose GROVE (Graph-based Routing Over Variant Experts), an adaptive framework that routes each example to SFT or RL based on the structural topology of its reasoning trace. GROVE parses each trace into a Reasoning Tree, scores its complexity via graph entropy, and partitions the corpus through a data-driven Gaussian Mixture Model with no hand-tuned threshold. Low-complexity traces are trained with SFT and high-complexity traces with GRPO as parallel LoRA adapters from a shared frozen base, merged into a single deployable model with no inference-time overhead. Across two training corpora and five reasoning benchmarks, GROVE consistently outperforms single-stage and naive-hybrid baselines, and a reversed-routing ablation confirms that this mapping is intrinsic rather than arbitrary.</abstract>
<identifier type="citekey">luo-etal-2026-grove</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1332</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1332/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>28210</start>
<end>28224</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T GROVE: Graph-based Routing Over Variant Experts for Adaptive Two-Stage Training of Reasoning Models
%A Luo, Yiming
%A Tian, Chunlin
%A Li, Li
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F luo-etal-2026-grove
%X Post-training reasoning models is computationally expensive, and the allocation between supervised fine-tuning (SFT) and reinforcement learning (RL) substantially affects reasoning quality. Existing methods operate at the training-procedure level and treat the reasoning corpus as homogeneous, ignoring the structural heterogeneity of individual traces. We propose GROVE (Graph-based Routing Over Variant Experts), an adaptive framework that routes each example to SFT or RL based on the structural topology of its reasoning trace. GROVE parses each trace into a Reasoning Tree, scores its complexity via graph entropy, and partitions the corpus through a data-driven Gaussian Mixture Model with no hand-tuned threshold. Low-complexity traces are trained with SFT and high-complexity traces with GRPO as parallel LoRA adapters from a shared frozen base, merged into a single deployable model with no inference-time overhead. Across two training corpora and five reasoning benchmarks, GROVE consistently outperforms single-stage and naive-hybrid baselines, and a reversed-routing ablation confirms that this mapping is intrinsic rather than arbitrary.
%R 10.18653/2026.findings-emnlp.1332
%U https://aclanthology.org/2026.findings-emnlp.1332/
%U https://doi.org/10.18653/2026.findings-emnlp.1332
%P 28210-28224
Markdown (Informal)
[GROVE: Graph-based Routing Over Variant Experts for Adaptive Two-Stage Training of Reasoning Models](https://aclanthology.org/2026.findings-emnlp.1332/) (Luo et al., Findings 2026)
ACL