Keeping Code-Aware LLMs Fresh: Full Refresh, In-Context Deltas, and Incremental Fine-Tuning

• URL: http://arxiv.org/abs/2511.14022v1
• Date: Tue, 18 Nov 2025 01:01:56 GMT
• Title: Keeping Code-Aware LLMs Fresh: Full Refresh, In-Context Deltas, and Incremental Fine-Tuning
• Abstract summary: We study how to keep a model fresh without surrendering retention on earlier code.<n>We compare three families of update strategies: (A) Full Refresh, retraining the entire model at the new snapshot; (B) In-Context Learning (ICL) that injects recent deltas at inference; and (C) Incremental Fine-Tuning (Inc-FT) on delta-derived training sets.<n>We contribute an alias-aware evaluation protocol that credits rename while never rewarding deleted paths, and a practical Forgetting Probe that quantifies residual emissions of obsolete paths.
• Score: 0.0
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Modern codebases evolve continuously: files are renamed or deleted; public APIs drift; behavior shifts within otherwise familiar modules. A model trained yesterday to map a developer's natural-language question to the exact set of repository file paths that matter will degrade tomorrow, even if the questions themselves look unchanged. In this paper we study, at system scale and across several widely used repositories, how to keep such a model fresh without surrendering retention on earlier code. We frame freshness as a form of domain drift between a base snapshot and the current HEAD, and we compare three families of update strategies: (A) Full Refresh, retraining the entire model at the new snapshot; (B) In-Context Learning (ICL) that injects recent deltas (raw git diffs or concise English summaries) at inference; and (C) Incremental Fine-Tuning (Inc-FT) on delta-derived training sets, with carefully controlled NEW:OLD mixing to mitigate catastrophic forgetting. We contribute an alias-aware evaluation protocol that credits rename while never rewarding deleted paths, and a practical Forgetting Probe that quantifies residual emissions of obsolete paths. Across Flask, SQLAlchemy, Pandas, and Poetry, Inc-FT with old-aware mixes delivers the best overall balance on mixed sets, ICL with English delta summaries delivers the fastest new-code lift when training is not feasible, and Full Refresh remains the ceiling when maximum NEW accuracy matters. We also compare Git-diff Inc-FT to full-file Inc-FT, showing that diffs excel in rename/delete-heavy windows while full-file context wins in behavior-change-heavy windows.

Related papers

• In-Cell Learning: Deployed Language Models Can Learn New Knowledge Without Changing a Single Stored Bit [4.726337413307433]
  We show that a 4-bit release stores each weight as an integer code and a shared scale, and leaves between the stored values an interval the quantizer discards.<n>We call in-cell learning the paradigm in which new knowledge is written only into that interval.<n>CellFill realizes the paradigm by construction: a low-rank position inside the cell, trained on the vendor's own 4-bit release.
  arXiv  Detail & Related papers  (2026-08-21T08:49:07Z)
• CORVUS: Context Optimization and Reduction Via Underlying Synchronization for LLM Coding Agents [8.604672064317235]
  CORVUS is a novel trajectory architecture that decouples file-read actions from their observations.<n>It produces significantly lighter-weight trajectories that remain synchronized with the actual state by construction.
  arXiv  Detail & Related papers  (2026-07-20T23:01:39Z)
• SEDCoT: Enhancing LLM-Based COBOL Code Translation via Symbolic Execution and Delta Debugging [49.898541267264115]
  Maintenance is increasingly challenging due to outdated technologies, sparse documentation, and developer retirement.<n>Traditional rule-based transcompilers yield outputs that are difficult to read and maintain.<n>We propose SEDCoT, a novel-to-C translation framework.
  arXiv  Detail & Related papers  (2026-07-05T03:07:10Z)
• REVES: REvision and VErification--Augmented Training for Test-Time Scaling [53.197756110943395]
  We propose a two-stage iterative framework that alternates between online data/prompt augmentation and policy optimization.<n>We observe gains of +6.5 points over the RL baseline and +4.0 points over standard multi-turn training.
  arXiv  Detail & Related papers  (2026-06-17T10:37:23Z)
• A Data-Efficient Path to Multilingual LLMs: Language Expansion via Post-training PARAM$Δ$ Integration into Upcycled MoE [61.0405668388095]
  We introduce method, which upcycles a dense model into a Mixture-of-Experts(MoE) architecture, allocating different experts to different languages.<n>Experiments demonstrate method's superiority even against baselines with similar FLOPs or number of parameters.
  arXiv  Detail & Related papers  (2026-05-18T08:59:08Z)
• Learning to Commit: Generating Organic Pull Requests via Online Repository Memory [11.042326503752756]
  Large language model (LLM)-based coding agents achieve impressive results on controlled benchmarks yet routinely produce pull requests that real maintainers reject.<n>We introduce Learning to Commit, a framework that closes this gap through Online Repository Memory.
  arXiv  Detail & Related papers  (2026-03-27T17:58:56Z)
• Pull Requests as a Training Signal for Repo-Level Code Editing [49.82435173554125]
  Clean Pull Request (Clean-PR) is a mid-training paradigm that leverages real-world GitHub pull requests as a training signal for repository-level editing.<n>We introduce a scalable pipeline that converts noisy pull request diffs into Search/Replace edit blocks through reconstruction and validation.<n>On SWE-bench, our model significantly outperforms the instruction-tuned baseline, achieving absolute improvements of 13.6% on SWE-bench Lite and 12.3% on SWE-bench Verified.
  arXiv  Detail & Related papers  (2026-02-07T09:22:25Z)
• Refinement Provenance Inference: Detecting LLM-Refined Training Prompts from Model Behavior [58.751981587234916]
  This paper formalizes the Refinement Provenance Inference (RPI) audit task as Refinement Provenance Inference (RPI)<n>We propose RePro, a logit-based framework that fuses teacher-forced likelihood features with logit-ranking signals.<n>During training, RePro learns a transferable representation via shadow fine-tuning, and uses a lightweight linear head to infer provenance on unseen victims without training-data access.
  arXiv  Detail & Related papers  (2026-01-05T10:16:41Z)
• Training-Free Time Series Classification via In-Context Reasoning with LLM Agents [29.14242392533328]
  Time series classification (TSC) spans diverse application scenarios, yet labeled data are often scarce.<n>We propose FETA, a multi-agent framework for training-free TSC via exemplar-based in-context reasoning.
  arXiv  Detail & Related papers  (2025-10-07T14:07:43Z)
• Impact-driven Context Filtering For Cross-file Code Completion [19.717276276962483]
  Retrieval-augmented generation (RAG) has recently demonstrated considerable potential for repository-level code completion.<n>To better understand the contribution of the retrieved cross-file contexts, we evaluate the impact of each retrieved code chunk on the completion.<n>These results underscore the potential of CODE to enhance the accuracy, efficiency, and attributability of repository-level code completion.
  arXiv  Detail & Related papers  (2025-08-08T03:08:19Z)
• Query Drift Compensation: Enabling Compatibility in Continual Learning of Retrieval Embedding Models [18.67755359053732]
  We study how the already indexed corpus can still be effectively used without the need of re-indexing.<n>We employ embedding distillation on both query and document embeddings to maintain stability.<n>We propose a novel query drift compensation method during retrieval to project new model query embeddings to the old embedding space.
  arXiv  Detail & Related papers  (2025-05-27T14:52:52Z)
• A Pipeline of Augmentation and Sequence Embedding for Classification of Imbalanced Network Traffic [0.0]
  We propose a pipeline to balance the dataset and classify it using a robust and accurate embedding technique.<n>We demonstrate that the proposed augmentation pipeline, combined with FS-Embedding, increases convergence speed and leads to a significant reduction in the number of model parameters.
  arXiv  Detail & Related papers  (2025-02-26T07:55:24Z)
• RAFT: Adapting Language Model to Domain Specific RAG [75.63623523051491]
  We present Retrieval Augmented FineTuning (RAFT), a training recipe that improves the model's ability to answer questions in a "openbook" in-domain settings. RAFT accomplishes this by citing the verbatim right sequence from the relevant document that would help answer the question. RAFT consistently improves the model's performance across PubMed, HotpotQA, and Gorilla datasets.
  arXiv  Detail & Related papers  (2024-03-15T09:26:02Z)
• RepoFusion: Training Code Models to Understand Your Repository [12.621282610983592]
  Large Language Models (LLMs) in coding assistants like GitHub Copilot struggle to understand the context present in the repository. Recent work has shown the promise of using context from the repository during inference. We propose RepoFusion, a framework to train models to incorporate relevant repository context.
  arXiv  Detail & Related papers  (2023-06-19T15:05:31Z)
• Consistent Diffusion Models: Mitigating Sampling Drift by Learning to be Consistent [97.64313409741614]
  We propose to enforce a emphconsistency property which states that predictions of the model on its own generated data are consistent across time. We show that our novel training objective yields state-of-the-art results for conditional and unconditional generation in CIFAR-10 and baseline improvements in AFHQ and FFHQ.
  arXiv  Detail & Related papers  (2023-02-17T18:45:04Z)
• BAFFLE: A Baseline of Backpropagation-Free Federated Learning [71.09425114547055]
  Federated learning (FL) is a general principle for decentralized clients to train a server model collectively without sharing local data. We develop backpropagation-free federated learning, dubbed BAFFLE, in which backpropagation is replaced by multiple forward processes to estimate gradients. BAFFLE is 1) memory-efficient and easily fits uploading bandwidth; 2) compatible with inference-only hardware optimization and model quantization or pruning; and 3) well-suited to trusted execution environments.
  arXiv  Detail & Related papers  (2023-01-28T13:34:36Z)
• Queried Unlabeled Data Improves and Robustifies Class-Incremental Learning [133.39254981496146]
  Class-incremental learning (CIL) suffers from the notorious dilemma between learning newly added classes and preserving previously learned class knowledge. We propose to leverage "free" external unlabeled data querying in continual learning. We show queried unlabeled data can continue to benefit, and seamlessly extend CIL-QUD into its robustified versions.
  arXiv  Detail & Related papers  (2022-06-15T22:53:23Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.