Lightning Weave: Improving the Accuracy-Efficiency Frontier of Reasoning Models through Capability Composition

Hugging Face Daily Papers Papers
reasoning-models capability-composition on-policy-distillation accuracy-efficiency math-benchmarks code-benchmarks

Summary

This paper introduces Lightning Weave, a framework that combines separately trained reasoning capabilities into one efficient model using on-policy distillation, enhancing accuracy and reducing token usage in math and code tasks.

A core goal of efficient reasoning is to improve the accuracy-efficiency frontier. However, jointly improving reasoning accuracy and inference efficiency can be challenging, as the two objectives can favor different reasoning behaviors. Independently post-trained models already offer distinct strengths in accuracy and efficiency. We introduce Lightning Weave, a post-training framework that extracts and composes these independently learned capabilities in a single student through on-policy distillation. Each acquired capability is represented by the policy shift from the model before post-training to the resulting specialist. Lightning Weave combines aligned log-ratio shifts at shared student token states and uses Tilted-Target DOPD to convert the cached signals into a stable learning target. Each anchor pair scores the cached trajectories once, enabling subsequent student training without serving multiple live anchor models concurrently. Across diverse student models and benchmarks in mathematics and code, Lightning Weave substantially improves upon the base students and achieves a state-of-the-art accuracy-efficiency frontier. On Qwen3.5-4B, it raises HMMT 2025 accuracy from 59.2% to 64.0% with 10.7% fewer response tokens, and LiveCodeBench v5 accuracy from 41.7% to 54.2% with 9.6% fewer response tokens. Adjusting the relative strengths of the anchor signals yields a strong empirical accuracy-efficiency Pareto frontier. These results establish Lightning Weave as a new practical route to efficient reasoning through capability composition. Code will be released soon.
Original Article
View Cached Full Text

Cached at: 09/16/26, 02:45 AM

Paper page - Lightning Weave: Improving the Accuracy-Efficiency Frontier of Reasoning Models through Capability Composition

Source: https://huggingface.co/papers/2609.14708

Abstract

Lightning Weave composes independently trained reasoning capabilities into a single efficient student via on-policy distillation, improving both accuracy and token efficiency across math and code benchmarks.

A core goal of efficient reasoning is to improve theaccuracy-efficiency frontier. However, jointly improving reasoning accuracy and inference efficiency can be challenging, as the two objectives can favor different reasoning behaviors. Independently post-trained models already offer distinct strengths in accuracy and efficiency. We introduce Lightning Weave, a post-training framework that extracts and composes these independently learned capabilities in a single student throughon-policy distillation. Each acquired capability is represented by thepolicy shiftfrom the model before post-training to the resulting specialist. Lightning Weave combines alignedlog-ratio shiftsat shared student token states and usesTilted-Target DOPDto convert the cached signals into a stable learning target. Each anchor pair scores the cached trajectories once, enabling subsequent student training without serving multiple live anchor models concurrently. Across diverse student models and benchmarks in mathematics and code, Lightning Weave substantially improves upon the base students and achieves a state-of-the-artaccuracy-efficiency frontier. On Qwen3.5-4B, it raises HMMT 2025 accuracy from 59.2% to 64.0% with 10.7% fewer response tokens, and LiveCodeBench v5 accuracy from 41.7% to 54.2% with 9.6% fewer response tokens. Adjusting the relative strengths of the anchor signals yields a strong empirical accuracy-efficiency Pareto frontier. These results establish Lightning Weave as a new practical route to efficient reasoning throughcapability composition. Code will be released soon.

View arXiv pageView PDFAdd to collection

Community

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.14708 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.14708 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.14708 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Improving Reasoning Capabilities in Small Models through Mixture-of-Layers Distillation with Stepwise Attention on Key Information

arXiv cs.CL

This paper proposes a novel Chain-of-Thought distillation framework that transfers teacher models' stepwise attention on key information to student models through a Mixture-of-Layers module for dynamic layer alignment. The method achieves consistent performance improvements on mathematical and commonsense reasoning benchmarks by explicitly guiding student models to progressively focus on critical information during reasoning.

OmniThoughtVis: A Scalable Distillation Pipeline for Deployable Multimodal Reasoning Models

arXiv cs.CL

This paper introduces OmniThoughtVis, a scalable pipeline for distilling multimodal reasoning capabilities from large teacher models to smaller, deployment-oriented MLLMs. The method uses curated chain-of-thought data to significantly improve reasoning performance on benchmarks like MathVerse and MMMU-Pro for models ranging from 2B to 8B parameters.