Teaching and Evaluating LLMs to Reason About Polymer Design Related Tasks

2026-01-22

AI Summary

This work addresses the lack of domain-specific expertise and inadequate alignment of current large language models (LLMs) in polymer design. To bridge this gap, the authors introduce PolyBench, a novel benchmark comprising 125,000 structured reasoning tasks, enriched with over 13 million experimental and synthesis data points. They propose a knowledge-enhanced reasoning distillation approach that fine-tunes smaller language models (7B–14B parameters) through hierarchical task organization and chain-of-thought (CoT) distillation. This method substantially improves model alignment and generalization on specialized scientific tasks, outperforming open-source models of comparable scale on PolyBench and even surpassing state-of-the-art closed-source LLMs. Consistent performance gains are also demonstrated across other polymer-related benchmarks, highlighting the effectiveness and transferability of the proposed framework.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Research in AI4Science has shown promise in many science applications, including polymer design. However, current LLMs prove ineffective on this problem space because: (i) most models lack polymer-specific knowledge (ii) existing aligned models lack coverage of knowledge and capabilities relevant to polymer design. Addressing this, we introduce PolyBench, a large scale training and test benchmark dataset of more than 125K polymer design related tasks, leveraging a knowledge base of 13M+ data points obtained from experimental and synthetic sources to ensure broad coverage of polymers and their properties. For effective alignment using PolyBench, we introduce a knowledge-augmented reasoning distillation method that augments this dataset with structured CoT. Furthermore, tasks in PolyBench are organized from simple to complex analytical reasoning problems, enabling generalization tests and diagnostic probes across the problem space. Experiments show that small language models (SLMs), of 7B to 14B parameters, trained on PolyBench data outperform similar sized models, and even closed source frontier LLMs on PolyBench test dataset while demonstrating gains on other polymer benchmarks as well.

Problem

• polymer design
• large language models
• AI4Science
• domain-specific knowledge
• reasoning tasks

Innovation

• PolyBench
• knowledge-augmented reasoning distillation
• chain-of-thought (CoT)
• polymer design
• AI4Science