@inproceedings{wang-etal-2026-aligning-llms,
title = "Aligning {LLM}s for Crystal Material Design: Test-Time Reinforcement Learning and Model Merging",
author = "Wang, Pengyu and
Han, Yang and
Chen, Xin and
Yu, Kai and
Zhu, Tong and
Chen, Lu and
Chen, Bo",
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.2238/",
pages = "47898--47919",
ISBN = "979-8-89176-462-0",
abstract = "Conditional crystal generation requires Large Language Models (LLMs) to produce Crystal Strings with discrete and continuous variables while satisfying composition, symmetry, and property constraints. Desired regions are often sparse in supervised data, while retraining for each objective preference is costly. We present a verifier-grounded framework combining label-free Test-Time Reinforcement Learning (TTRL) and Model Merging (MM). TTRL performs one batch-level, partially transductive adaptation per property using composition, symmetry, and learned band-gap rewards, without target reference structures. Property-specific experts are interpolated to navigate bounded symmetry{--}band-gap trade-offs without retraining intermediate policies. On MP-20, TTRL improves symmetry control, band-gap accuracy, and compositional fidelity over High-Precision SFT and Standard RL, and is competitive with diffusion and preference-aligned baselines. In a held-out high-band-gap region, TTRL generalizes to test prompts unseen during adaptation, reducing the mean squared error between surrogate-predicted and target band gaps by 52{\%} relative to standard reinforcement learning under best-of-ten sampling. An independent band-gap predictor shows consistent gains, while density functional theory calculations further assess selected candidates. Similar gains with Qwen3.5-9B provide cross-backbone evidence. Model Merging recovers competitive two-objective trade-offs from two endpoint experts without additional RL, enabling efficient adaptation and preference control for structured scientific generation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-aligning-llms">
<titleInfo>
<title>Aligning LLMs for Crystal Material Design: Test-Time Reinforcement Learning and Model Merging</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pengyu</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xin</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kai</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lu</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bo</namePart>
<namePart type="family">Chen</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>Conditional crystal generation requires Large Language Models (LLMs) to produce Crystal Strings with discrete and continuous variables while satisfying composition, symmetry, and property constraints. Desired regions are often sparse in supervised data, while retraining for each objective preference is costly. We present a verifier-grounded framework combining label-free Test-Time Reinforcement Learning (TTRL) and Model Merging (MM). TTRL performs one batch-level, partially transductive adaptation per property using composition, symmetry, and learned band-gap rewards, without target reference structures. Property-specific experts are interpolated to navigate bounded symmetry–band-gap trade-offs without retraining intermediate policies. On MP-20, TTRL improves symmetry control, band-gap accuracy, and compositional fidelity over High-Precision SFT and Standard RL, and is competitive with diffusion and preference-aligned baselines. In a held-out high-band-gap region, TTRL generalizes to test prompts unseen during adaptation, reducing the mean squared error between surrogate-predicted and target band gaps by 52% relative to standard reinforcement learning under best-of-ten sampling. An independent band-gap predictor shows consistent gains, while density functional theory calculations further assess selected candidates. Similar gains with Qwen3.5-9B provide cross-backbone evidence. Model Merging recovers competitive two-objective trade-offs from two endpoint experts without additional RL, enabling efficient adaptation and preference control for structured scientific generation.</abstract>
<identifier type="citekey">wang-etal-2026-aligning-llms</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2238/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>47898</start>
<end>47919</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Aligning LLMs for Crystal Material Design: Test-Time Reinforcement Learning and Model Merging
%A Wang, Pengyu
%A Han, Yang
%A Chen, Xin
%A Yu, Kai
%A Zhu, Tong
%A Chen, Lu
%A Chen, Bo
%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 wang-etal-2026-aligning-llms
%X Conditional crystal generation requires Large Language Models (LLMs) to produce Crystal Strings with discrete and continuous variables while satisfying composition, symmetry, and property constraints. Desired regions are often sparse in supervised data, while retraining for each objective preference is costly. We present a verifier-grounded framework combining label-free Test-Time Reinforcement Learning (TTRL) and Model Merging (MM). TTRL performs one batch-level, partially transductive adaptation per property using composition, symmetry, and learned band-gap rewards, without target reference structures. Property-specific experts are interpolated to navigate bounded symmetry–band-gap trade-offs without retraining intermediate policies. On MP-20, TTRL improves symmetry control, band-gap accuracy, and compositional fidelity over High-Precision SFT and Standard RL, and is competitive with diffusion and preference-aligned baselines. In a held-out high-band-gap region, TTRL generalizes to test prompts unseen during adaptation, reducing the mean squared error between surrogate-predicted and target band gaps by 52% relative to standard reinforcement learning under best-of-ten sampling. An independent band-gap predictor shows consistent gains, while density functional theory calculations further assess selected candidates. Similar gains with Qwen3.5-9B provide cross-backbone evidence. Model Merging recovers competitive two-objective trade-offs from two endpoint experts without additional RL, enabling efficient adaptation and preference control for structured scientific generation.
%U https://aclanthology.org/2026.findings-emnlp.2238/
%P 47898-47919
Markdown (Informal)
[Aligning LLMs for Crystal Material Design: Test-Time Reinforcement Learning and Model Merging](https://aclanthology.org/2026.findings-emnlp.2238/) (Wang et al., Findings 2026)
ACL