# SRA: Span Representation Alignment for Large Language Model Distillation

arXiv: https://arxiv.org/abs/2605.01205 · HTML: https://mlsanity.com/paper/2605.01205/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in LLM preference optimization methods, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 17 of its 29 references that are in the map.
- First posted on arXiv 2 May 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [LLM preference optimization methods](https://mlsanity.com/alignment/llm-preference-optimization-methods/)
- Knowledge distillation / [Knowledge distillation for LLMs](https://mlsanity.com/distill/knowledge-distillation-for-llms/)

## Builds on (most cited first, 17 of 17)

- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://mlsanity.com/paper/1910.01108/) (2019)
- [Neural Ordinary Differential Equations](https://mlsanity.com/paper/1806.07366/) (2018)
- [OPT: Open Pre-trained Transformer Language Models](https://mlsanity.com/paper/2205.01068/) (2022)
- [Self-Instruct: Aligning Language Models with Self-Generated Instructions](https://mlsanity.com/paper/2212.10560/) (2022)
- [TinyBERT: Distilling BERT for Natural Language Understanding](https://mlsanity.com/paper/1909.10351/) (2019)
- [Sequence-Level Knowledge Distillation](https://mlsanity.com/paper/1606.07947/) (2016)
- [TinyLlama: An Open-Source Small Language Model](https://mlsanity.com/paper/2401.02385/) (2024)
- [Patient Knowledge Distillation for BERT Model Compression](https://mlsanity.com/paper/1908.09355/) (2019)
- [DialogSum: A Real-Life Scenario Dialogue Summarization Dataset](https://mlsanity.com/paper/2105.06762/) (2021)
- [f-Divergence Minimization for Sequence-Level Knowledge Distillation](https://mlsanity.com/paper/2307.15190/) (2023)
- [Understanding and Improving Transformer From a Multi-Particle Dynamic System Point of View](https://mlsanity.com/paper/1906.02762/) (2019)
- [Towards Cross-Tokenizer Distillation: the Universal Logit Distillation Loss for LLMs](https://mlsanity.com/paper/2402.12030/) (2024)
- [Identifying and Mitigating Vulnerabilities in LLM-Integrated Applications](https://mlsanity.com/paper/2311.16153/) (2023)
- [Universal Cross-Tokenizer Distillation via Approximate Likelihood Matching](https://mlsanity.com/paper/2503.20083/) (2025)
- [Multi-Level Optimal Transport for Universal Cross-Tokenizer Knowledge Distillation on Language Models](https://mlsanity.com/paper/2412.14528/) (2024)
- [DWA-KD: Dual-Space Weighting and Time-Warped Alignment for Cross-Tokenizer Knowledge Distillation](https://mlsanity.com/paper/2602.21669/) (2026)

## Built on by (most cited first, 3 of 3)

- [TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching](https://mlsanity.com/paper/2605.12288/) (2026)
- [HyperProve: Answer-Guided Hypergraph Expansion for Multi-Hop Question Answering](https://mlsanity.com/paper/2609.13768/) (2026)
- [Dual-Vocabulary Language Model for Cross-Tokenizer Distillation](https://mlsanity.com/paper/2609.33816/) (2026)

## Abstract (arXiv)

Cross-Tokenizer Knowledge Distillation (CTKD) enables knowledge transfer between a large language model and a smaller student, even when they employ different tokenizers. While existing approaches mainly focus on token-level alignment strategies, which are often brittle and sensitive to discrepancies between tokenizers, we argue that the method of aggregating tokens into more robust representations before distillation is of equal importance. In this paper, we introduce SRA (Span Representation Alignment for Large Language Model Distillation), a novel framework that reframes CTKD through the physical lens of Multi-Particle Dynamical Systems. SRA shifts the fundamental unit of alignment from tokens to robust, tokenizer-agnostic spans. We model each span as a cluster of particles and represent its state by its Center of Mass (CoM) - an attention-weighted average that captures rich semantic information. We leverage the concept of span centers of mass with attention-derived weighting to prioritize the most salient spans. In addition, we employ a geometric regularizer to preserve the structural integrity of the representation space and introduce aligned span logit distillation to enhance knowledge transfer across models. In challenging cross-architecture distillation experiments, SRA consistently and significantly outperforms state-of-the-art CTKD baselines, validating our physically-grounded approach.