Lightweighting Diffusion Transformer and Sparse Attention Architectures with Tool Orchestration for Enhanced Multimodal Agent Reasoning
Real-time Streaming Implementation Based on Long-Context Understanding and Reinforcement Learning Optimization, and Full-Lifecycle Benchmark System Establishment

From Code Foundation Models to Agents and Applications: A Practical Guide to Code Intelligence

https://arxiv.org/abs/2511.18538

This paper is a practical guide systematically analyzing the complete lifecycle of code LLMs from data collection, model training, and fine-tuning through reinforcement learning to agent construction. It comparatively analyzes performance and design tradeoffs between general-purpose LLMs and code-specialized LLMs, and illuminates the gap between simple benchmark scores and actual software development environments. Through diverse experiments, it validates model scaling laws and hyperparameter sensitivity, presenting concrete methodologies and future research directions for applying academic research results to actual industrial settings.