# MUSCLE: A Model Update Strategy for Compatible LLM Evolution

arXiv: https://arxiv.org/abs/2407.09435 · HTML: https://mlsanity.com/paper/2407.09435/

## Key facts

- Cited by 15 papers in the map (data through 14 Oct 2026).
- Builds on 24 of its 47 references that are in the map.
- First posted on arXiv 12 Jul 2024.

## Builds on (most cited first, 24 of 24)

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [PIQA: Reasoning about Physical Commonsense in Natural Language](https://mlsanity.com/paper/1911.11641/) (2019)
- [SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems](https://mlsanity.com/paper/1905.00537/) (2019)
- [Crowdsourcing Multiple Choice Science Questions](https://mlsanity.com/paper/1707.06209/) (2017)
- [SAMSum Corpus: A Human-annotated Dialogue Dataset for Abstractive Summarization](https://mlsanity.com/paper/1911.12237/) (2019)
- [SWAG: A Large-Scale Adversarial Dataset for Grounded Commonsense Inference](https://mlsanity.com/paper/1808.05326/) (2018)
- [Towards Backward-Compatible Representation Learning](https://mlsanity.com/paper/2003.11942/) (2020)
- [ELLE: Efficient Lifelong Pre-training for Emerging Data](https://mlsanity.com/paper/2203.06311/) (2022)
- [Positive-Congruent Training: Towards Regression-Free Model Updates](https://mlsanity.com/paper/2011.09161/) (2020)
- [Self-supervised Knowledge Distillation for Few-shot Learning](https://mlsanity.com/paper/2006.09785/) (2020)
- [An Empirical Analysis of Backward Compatibility in Machine Learning Systems](https://mlsanity.com/paper/2008.04572/) (2020)
- [(Why) Is My Prompt Getting Worse? Rethinking Regression Testing for Evolving LLM APIs](https://mlsanity.com/paper/2311.11123/) (2023)
- [Forward Compatible Training for Large-Scale Embedding Retrieval Systems](https://mlsanity.com/paper/2112.02805/) (2021)
- [Fantastic Gains and Where to Find Them: On the Existence and Prospect of General Knowledge Transfer between Any Pretrained Model](https://mlsanity.com/paper/2310.17653/) (2023)
- [X-Adapter: Adding Universal Compatibility of Plugins for Upgraded Diffusion Model](https://mlsanity.com/paper/2312.02238/) (2023)
- [Recyclable Tuning for Continual Pre-training](https://mlsanity.com/paper/2305.08702/) (2023)
- [Backward-Compatible Prediction Updates: A Probabilistic Approach](https://mlsanity.com/paper/2107.01057/) (2021)
- [Regression Bugs Are In Your Model! Measuring, Reducing and Analyzing Regressions In NLP Model Updates](https://mlsanity.com/paper/2105.03048/) (2021)
- [Measuring and Reducing Model Update Regression in Structured Prediction for NLP](https://mlsanity.com/paper/2202.02976/) (2022)
- [ELODI: Ensemble Logit Difference Inhibition for Positive-Congruent Training](https://mlsanity.com/paper/2205.06265/) (2022)
- [Improving Prediction Backward-Compatiblility in NLP Model Upgrade with Gated Fusion](https://mlsanity.com/paper/2302.02080/) (2023)

## Built on by (most cited first, 15 of 15)

- [Eureka: Evaluating and Understanding Large Foundation Models](https://mlsanity.com/paper/2409.10566/) (2024)
- [Efficient Construction of Model Family through Progressive Training Using Model Expansion](https://mlsanity.com/paper/2504.00623/) (2025)
- [boldsymbolλ-Orthogonality Regularization for Compatible Representation Learning](https://mlsanity.com/paper/2509.16664/) (2025)
- [Efficient Model Development through Fine-tuning Transfer](https://mlsanity.com/paper/2503.20110/) (2025)
- [Beyond the Mean: Within-Model Reliable Change Detection for LLM Evaluation](https://mlsanity.com/paper/2604.27405/) (2026)
- [LoRASuite: Efficient LoRA Adaptation Across Large Language Model Upgrades](https://mlsanity.com/paper/2505.13515/) (2025)
- [Mitigating Negative Flips via Margin Preserving Training](https://mlsanity.com/paper/2511.08322/) (2025)
- [Test Before You Deploy: Governing Updates in the LLM Supply Chain](https://mlsanity.com/paper/2604.27789/) (2026)
- [Same Agent, Different Answers: A Repeat-Aware Audit of Corpus-Induced Answer Churn in Retrieval-Augmented QA](https://mlsanity.com/paper/2608.22856/) (2026)
- [Certifying Model Upgrades with Slice-Wise Non-Regression and Incumbent Fallback](https://mlsanity.com/paper/2609.13714/) (2026)
- [LLM Evolution as an Industry-Scale Ecosystem: A Lifecycle Perspective on Continual Learning](https://mlsanity.com/paper/2606.24901/) (2026)
- [No Universal Signal Predicts Sample-Level LLM Regression under Version Updates](https://mlsanity.com/paper/2608.13607/) (2026)
- [What Aggregate Scores Miss: Measuring Item-Level Regressions in Commercial LLM API Migrations](https://mlsanity.com/paper/2608.17719/) (2026)
- [Learn Here, Move Less Elsewhere: Input-Conditioned Plasticity from Retained-Domain Activation Atlases](https://mlsanity.com/paper/2609.34478/) (2026)
- [Spherical Interpolation for Backward-Compatible Multimodal Representations](https://mlsanity.com/paper/2609.39836/) (2026)

## Abstract (arXiv)

Large Language Models (LLMs) are regularly updated to enhance performance, typically through changes in data or architecture. Within the update process, developers often prioritize improving overall performance metrics, paying less attention to maintaining compatibility with earlier model versions. Instance-level degradation (instance regression) of performance from one model version to the next can interfere with a user's mental model of the capabilities of a particular language model. Users having to adapt their mental model with every update can lead to dissatisfaction, especially when the new model has degraded compared to a prior version for a known use case (model update regression). We find that when pretrained LLM base models are updated, fine-tuned user-facing downstream task adapters experience negative flips -- previously correct instances are now predicted incorrectly. We observe model update regression between different model versions on a diverse set of tasks and models, even when the downstream task training procedures remain identical. We argue for the importance of maintaining model update compatibility during updates, and present evaluation metrics designed specifically for generative tasks, while also being applicable to discriminative tasks. We propose a training strategy to minimize the extent of instance regression in model updates, involving training of a compatibility adapter that can enhance task fine-tuned language models. We show negative flips reduce by up to 40% e.g. when updating Llama 1 to Llama 2 with our proposed method.