# Memory-Efficient Large Language Models for Program Repair with Semantic-Guided Patch Generation

arXiv: https://arxiv.org/abs/2410.16655 · HTML: https://mlsanity.com/paper/2410.16655/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 39 of its 81 references that are in the map.
- First posted on arXiv 22 Oct 2024.

## Builds on (most cited first, 30 of 39)

- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [The Pile: An 800GB Dataset of Diverse Text for Language Modeling](https://mlsanity.com/paper/2101.00027/) (2020)
- [Code Llama: Open Foundation Models for Code](https://mlsanity.com/paper/2308.12950/) (2023)
- [Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation](https://mlsanity.com/paper/2305.01210/) (2023)
- [Qwen2.5-Coder Technical Report](https://mlsanity.com/paper/2409.12186/) (2024)
- [BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions](https://mlsanity.com/paper/2406.15877/) (2024)
- [InCoder: A Generative Model for Code Infilling and Synthesis](https://mlsanity.com/paper/2204.05999/) (2022)
- [AutoCodeRover: Autonomous Program Improvement](https://mlsanity.com/paper/2404.05427/) (2024)
- [Reinforcement learning](https://mlsanity.com/paper/2405.10369/) (2024)
- [RepairAgent: An Autonomous, LLM-Based Agent for Program Repair](https://mlsanity.com/paper/2403.17134/) (2024)
- [Using an LLM to Help With Code Understanding](https://mlsanity.com/paper/2307.08177/) (2023)
- [Planning with Large Language Models for Code Generation](https://mlsanity.com/paper/2303.05510/) (2023)
- [SequenceR: Sequence-to-Sequence Learning for End-to-End Program Repair](https://mlsanity.com/paper/1901.01808/) (2018)
- [Few-shot training LLMs for project-specific code-summarization](https://mlsanity.com/paper/2207.04237/) (2022)
- [Keep the Conversation Going: Fixing 162 out of 337 bugs for 0.42 each using ChatGPT](https://mlsanity.com/paper/2304.00385/) (2023)
- [InferFix: End-to-End Program Repair with LLMs](https://mlsanity.com/paper/2303.07263/) (2023)
- [An Empirical Study on Learning Bug-Fixing Patches in the Wild via Neural Machine Translation](https://mlsanity.com/paper/1812.08693/) (2018)
- [CURE: Code-Aware Neural Machine Translation for Automatic Program Repair](https://mlsanity.com/paper/2103.00073/) (2021)
- [Less Training, More Repairing Please: Revisiting Automated Program Repair via Zero-shot Learning](https://mlsanity.com/paper/2207.08281/) (2022)
- [TBar: Revisiting Template-based Automated Program Repair](https://mlsanity.com/paper/1903.08409/) (2019)
- [A Syntax-Guided Edit Decoder for Neural Program Repair](https://mlsanity.com/paper/2106.08253/) (2021)
- [Copiloting the Copilots: Fusing Large Language Models with Completion Engines for Automated Program Repair](https://mlsanity.com/paper/2309.00608/) (2023)
- [Neural Program Repair with Execution-based Backpropagation](https://mlsanity.com/paper/2105.04123/) (2021)
- [RepairLLaMA: Efficient Representations and Fine-Tuned Adapters for Program Repair](https://mlsanity.com/paper/2312.15698/) (2023)
- [CoditT5: Pretraining for Source Code and Natural Language Editing](https://mlsanity.com/paper/2208.05446/) (2022)
- [Getafix: Learning to Fix Bugs Automatically](https://mlsanity.com/paper/1902.06111/) (2019)
- [SelfAPR: Self-supervised Program Repair with Test Execution Diagnostics](https://mlsanity.com/paper/2203.12755/) (2022)
- [GAMMA: Revisiting Template-based Automated Program Repair via Mask Prediction](https://mlsanity.com/paper/2309.09308/) (2023)
- [Trust Enhancement Issues in Program Repair](https://mlsanity.com/paper/2108.13064/) (2021)
- [Practical Program Repair via Bytecode Mutation](https://mlsanity.com/paper/1807.03512/) (2018)

## Built on by (most cited first, 5 of 5)

- [From Code Foundation Models to Agents and Applications: A Comprehensive Survey and Practical Guide to Code Intelligence](https://mlsanity.com/paper/2511.18538/) (2025)
- [PAGENT: Learning to Patch Software Engineering Agents](https://mlsanity.com/paper/2506.17772/) (2025)
- [Perish or Flourish? A Holistic Evaluation of Large Language Models for Code Generation in Functional Programming](https://mlsanity.com/paper/2601.02060/) (2026)
- [FPMoE: A Sparse Mixture-of-Experts Approach to Functional Code Generation](https://mlsanity.com/paper/2605.27849/) (2026)
- [How Do LLMs Read Bug Reports? An Empirical Study of Attention in LLMs for Automated Program Repair](https://mlsanity.com/paper/2607.25873/) (2026)

## Abstract (arXiv)

In this paper, we first show that increases in beam size, even for small-sized LLMs (1B-7B params), require extensive GPU usage, leading to up to 80% of recurring crashes due to memory overloads in LLM-based APR. Seemingly simple solutions to reduce memory consumption are (1) to quantize LLM models, i.e., converting the weights of an LLM from high-precision values to lower-precision ones, and (2) to make beam search sequential, i.e., forwarding each beam through the model sequentially and then concatenating them back into a single output. However, we show that these approaches still do not work via both theoretical analysis and experiments. To address this, we introduce FLAMES, a novel LLM-based APR technique that employs semantic-guided patch generation to enhance repair effectiveness and memory efficiency. Unlike conventional methods that rely on beam search, FLAMES utilizes greedy decoding to enhance memory efficiency while steering the search towards more potentially good repair candidates via a semantic-guided best-first search algorithm. At each decoding step, FLAMES uses semantic feedback from test validation, such as the number of passing and failing test cases, to select the most promising token to explore further. Our empirical evaluation on Defects4J shows thatFLAMES substantially reduces memory consumption by up to 83% compared to LLM-based APR without compromising time efficiency. Moreover, FLAMES correctly fixes 133 bugs on Defects4J, fixing 10 bugs more than the best baseline. Additionally, these improvements also generalize to the HumanEval-Java and TransformedD4J datasets, where FLAMES generates 12% and 36.5% more correct patches, respectively, than the best baseline.