프로그래밍 지식 그래프를 활용한 컨텍스트 기반 코드 생성: 검색 성능 향상 및 환각 현상 감소
대규모 언어 모델(LLM) 기반 코드 생성 성능을 향상시키기 위해, 본 논문에서는 프로그래밍 지식 그래프(PKG)를 활용한 새로운 컨텍스트 기반 코드 생성 프레임워크를 제안합니다. PKG는 코드 검색의 정확도를 높이고, 트리 가지치기 기법을 통해 관련 없는 정보를 줄여 환각 현상을 감소시킵니다.
Context-Aware Code Generation with Programming Knowledge Graphs: A Novel Approach
This research paper introduces a novel framework for enhancing code generation using Programming Knowledge Graphs (PKG) and a re-ranking mechanism to improve the accuracy and relevance of generated code.
MojoBench: A Framework for Evaluating and Improving Mojo Code Generation in Large Language Models
Large language models (LLMs) often struggle with code generation in emerging programming languages like Mojo. MojoBench introduces a novel framework, including a benchmark dataset and specialized LLMs, to evaluate and enhance Mojo code generation capabilities, highlighting the importance of domain-specific pretraining and targeted finetuning.
CONAN: A Retrieval-Augmented Language Model for Code Assistance
CONAN, a novel retrieval-augmented language model, effectively assists code generation, summarization, and completion by leveraging a structure-aware retriever and a dual-view code representation mechanism.
Meta's RLEF Achieves State-of-the-Art Coding Performance by Fine-Tuning LLMs with Execution Feedback
Meta's novel RLEF method significantly improves code generation capabilities of Large Language Models (LLMs) by leveraging execution feedback, achieving state-of-the-art results and surpassing even GPT-4 in efficiency and accuracy.
RTLCoder: An Open-Source LLM for Efficient RTL Code Generation
RTLCoder is a novel, open-source, and efficient large language model (LLM) specifically designed for generating RTL code from natural language instructions, outperforming existing commercial and open-source solutions in accuracy and efficiency.
Horizon-Length Prediction: Improving Code Generation by Predicting Completion Length
Current code language models struggle with accurately filling in missing code because they lack the ability to plan ahead. This paper introduces Horizon-Length Prediction (HLP), a novel training objective that teaches models to predict the length of the missing code, significantly improving their ability to generate coherent and accurate code completions.
Learning to Self-Refine Code Generation with Cycle Framework
Pre-trained code language models struggle with self-refinement, but Cycle framework enhances self-refinement capability, improving code generation performance.
Grounding Data Science Code Generation with Input-Output Specifications: Enhancing LLMs with Execution-Based Feedback
Enhancing large language models (LLMs) with execution-based feedback improves code generation accuracy for data science tasks.
© 2024 by Linnk AI