The field of large language models (LLMs) for code generation and optimization is rapidly evolving, with a focus on improving the accuracy, reliability, and efficiency of these models. Recent developments have explored the use of retrieval-augmented generation, multi-agent collaboration, and runtime debugging to enhance code generation functionality. Additionally, there is a growing interest in applying LLMs to specialized domains, such as high-performance computing and automated code optimization. Noteworthy papers in this area include CHORUS, which proposes a hierarchical retrieval-augmented generation framework for synthesizing linear programming code, and MARCO, which presents a multi-agent system for optimizing HPC code generation using LLMs. Other notable papers, such as AutoPatch and CompileAgent, demonstrate the potential of LLMs in patching vulnerable code and automating real-world repo-level compilation, respectively.
Advances in Large Language Models for Code Generation and Optimization
Sources
CHORUS: Zero-shot Hierarchical Retrieval and Orchestration for Generating Linear Programming Code
Enhancing LLM Code Generation: A Systematic Evaluation of Multi-Agent Collaboration and Runtime Debugging for Improved Accuracy, Reliability, and Latency
Iterative Resolution of Prompt Ambiguities Using a Progressive Cutting-Search Approach
MARCO: A Multi-Agent System for Optimizing HPC Code Generation Using Large Language Models
Identification and Optimization of Redundant Code Using Large Language Models
AutoPatch: Multi-Agent Framework for Patching Real-World CVE Vulnerabilities
CompileAgent: Automated Real-World Repo-Level Compilation with Tool-Integrated LLM-based Agent System
Towards Mitigating API Hallucination in Code Generated by LLMs with Hierarchical Dependency Aware