#code generation
41 papers match
The Case for Vibe Modeling: A Missing Step in AI-Based Trustworthy Software Development
Shalini Chakraborty, Michael Mittermaier, Judith Michael
The paper proposes "vibe modeling" as a lightweight intermediate abstraction between natural language prompts and code generation by large language models, aiming to improve unders…
CoGate: Confidence-Gated Co-Decoding for Secure Code Generation
Minghao Hu, Lannan Luo, Allen Roush +1
The paper introduces CoGate, a method that uses the confidence of a security expert model to gate its influence during co-decoding for generating more secure code with large langua…
IndustryForge-27B: A Domain-Enhanced Multimodal Foundation Model for Industrial CAD
Nianchen Deng, Jiaxin Ai, Tao Hu +10
The paper introduces IndustryForge-27B, a multimodal foundation model fine‑tuned on diverse industrial CAD data to understand drawings, generate parametric modeling scripts, and co…
Lightning OPD 2.0: Mitigating Style Bias in Cross-Teacher On-Policy Distillation for Large Reasoning Models
Yecheng Wu, Song Han, Han Cai
The paper proposes Lightning OPD 2.0, a method that reduces style‑related bias when using on‑policy distillation across different teacher models, improving performance on mathemati…
MRCoder: An Efficient Context Selecting Approach for Repository-Level Code Generation
Peiding Wang, Li Zhang, Fang Liu
The paper introduces MRCoder, a map-reduce based framework that selects and refines repository-specific code context using lightweight draft models, improving the accuracy and effi…
From Interface to Inference: Eliciting Any-Order Inference from Any-Order Models
Seunggeun Kim, Jaeyeon Kim, Taekyun Lee +4
The paper investigates how to give language models a native ability to reason and generate text in any order, introducing insertion‑based and latent‑space masked diffusion methods…
CodeSpec: Dual Executable Specifications for Agentic Long-Horizon Feature Development
Peiding Wang, Li Zhang, Fang Liu +2
The paper introduces CodeSpec, a method that creates paired architecture and behavior specifications to guide LLM-based code agents in developing new features across existing codeb…
DHRCL:Training Code LLMs with Dense Hierarchical Rewards and Curriculum Learning
Shuhang Wang, Ziming Li, Hui Cheng
The paper introduces DHRCL, a reinforcement‑learning framework for code‑focused large language models that uses a hierarchy of dense rewards (syntax, execution, unit‑test pass, and…
MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program Synthesis
Yihao Chen, Shi Chang, Khaled Chawa +4
MindForge automatically turns open‑source command‑line programs into source‑free environments that expose only compiled executables and documentation, enabling the training of smal…
Fewer Clarifications, Better Code: Benchmarking Cross-Session Personalized Ambiguity Adaptation in Coding Assistants
Zijian Xu, Wenshuo Zhang, Zisen Qin +4
The paper defines personalized ambiguity adaptation for coding assistants, introduces the CAPA benchmark to evaluate how well models use a user's past resolved sessions to handle r…
RLPF: Reinforcement Learning from Performance Feedback for Code Generation
Huihao Jing, Haozhe Cui, Wenbin Hu +9
The paper introduces RLPF, a reinforcement‑learning approach that uses staged performance feedback to train code‑generation models to produce not only correct programs but also fas…
VisualPatchWorld: Code World Models as Latent Structured Representations for Planning
Jiaxin Bai, Jiaxuan Xiong
The paper presents VisualPatchWorld, a system that learns compact code programs to model world dynamics from visual observations, enabling inspection, simulation, and use in model-…
Weak-to-Strong On-Policy Distillation
Fangxu Yu, Zinan Lin, Xiaodong Liu +4
The paper proposes Weak-to-Strong On-Policy Distillation (W2S-OPD), a method that improves a large language model by distilling knowledge from multiple weaker models using a constr…
TraceCoder: Explainable and Auditable Code Generation with Position-Key Snippet Versioning
Rwaida Alssadi, Muntaser Syed, Balaji Kasula +6
The paper introduces TraceCoder, a system that records detailed provenance for each code snippet generated by large language models, visualizes the evolution of code through repair…
Do Context Files Help Coding Agents? A Two-Agent Ablation Study on Real Repositories
Prakhar Khatri
The paper investigates whether providing persistent context files (like AGENTS.md) improves the performance of AI coding agents, finding no measurable benefit across Claude Code an…
Code Correctness Is Linearly Decodable from LLM Hidden States Before Generation
Carlo Di Cicco
The paper shows that the hidden state of a large language model right before it starts generating code contains a linear signal that predicts whether the produced code will be corr…
Large Language Models for Code Generation from Multilingual Prompts: A Curated Benchmark and a Study on Code Quality
Saima Afrin, Alessandro Midolo, Camilo Escobar-Velásquez +5
The paper introduces a curated multilingual benchmark to study how the natural language of prompts influences code generation quality of large language models, evaluating functiona…
Function-Aware Fill-in-the-Middle as Mid-Training for Coding Agent Foundation Models
Yubo Wang, Jiarong Liang, Yuxuan Zhang +5
The paper introduces a function-aware fill-in-the-middle (FIM) mid‑training method that masks function calls in code to improve coding agents' ability to incorporate tool outputs,…
Can LLMs Build a MaxSAT Solver from Papers? The CoreForge Experience
Ruben Martins
The paper describes CoreForge, an effort to use large language models to implement an unweighted MaxSAT solver directly from research papers, detailing the workflow, components bui…
Capturing and Exploiting Design Pattern Variability in Mobile Application Generation
Ramón Peralta, Jose-Miguel Horcas
The increasing reliance on automatic code generation in mobile application development often leads to code that neglects fundamental design principles and architectural quality. In…
LQCDMaster: Agentic Scientific Computing for Lattice Quantum Chromodynamics Research
Haofei Gao, Tingjia Miao, Wenkai Jin +12
LQCDMaster is an AI-driven scientific computing agent that translates natural‑language lattice QCD research tasks into fully executable PyQUDA workflows, automating code generation…
Token Reduction Is Not Cost Reduction
Sarel Weinberger, Amir Hozez
The paper studies whether context‑reduction techniques for API‑based coding agents actually lower the billed cost of using large language models, showing that token reduction often…
Rethinking the Capability of Fine-Tuned Language Models for Automated Vulnerability Repair
Woorim Han, Yeongjun Kwak, Miseon Yu +4
The paper investigates how well fine‑tuned language models can automatically fix unseen software vulnerabilities and critiques common token‑level evaluation metrics, introducing a…
Design-Specification Tiling for ICL-based CAD Code Generation
Yali Du, San-Zhuo Xi, Hui Sun +1
The paper introduces Design‑Specification Tiling, a method for selecting in‑context learning exemplars that maximizes coverage of CAD design requirements, using a knowledge‑suffici…
One search, two signals: results blend meaning (embedding similarity, so papers that never use your words still surface) with keyword matches on titles, abstracts and summaries. Free, no sign-in needed.