AmberLJC/LLMSys-PaperList

Large Language Model (LLM) Systems Paper List

What it solves

This project addresses the challenge of keeping up with the rapidly evolving field of Large Language Model (LLM) systems. It provides a centralized, curated repository of academic papers, technical reports, and projects to help researchers and engineers track developments in how LLMs are trained, served, and optimized.

How it works

The repository organizes a vast collection of research into logical categories such as Training (Pre-training and Post-training/RLHF), Serving, Multi-modal systems, and LLM-based systems. It includes a "Trends at a Glance" analysis that categorizes research by the "unit of work" being optimized—moving from individual requests in 2024 to reasoning traces in 2025 and full agent trajectories in 2026.

Who it’s for

It is designed for researchers, systems engineers, and students who need a structured way to navigate the booming landscape of LLM systems research and stay informed about the latest technical breakthroughs.

Highlights

  • Categorized Research: Extensive lists of papers covering pre-training, fault tolerance, serving, and agentic systems.
  • Trend Analysis: Visualizations and data showing the shift in research focus toward agentic systems, compiler/kernel design, and energy efficiency.
  • Historical Context: Papers organized by year, allowing users to trace the evolution of system optimizations from 2024 through 2026.
  • Broad Scope: Includes industrial technical reports, ML conference papers, and educational resources like course materials.

Related

  • Project
    zhaoyang97/Paper-Notes

    A comprehensive, community-driven repository of over 23,000 concise AI paper notes categorized by conference and research field to help users quickly grasp the latest industry advancements.

  • Project
    raphaelmansuy/edgequake

    EdgeQuake is a Rust‑based, single‑binary Graph‑RAG system that stores documents as a knowledge graph in PostgreSQL (pgvector + Apache AGE). It works with any OpenAI‑compatible LLM—cloud or local—via a simple Docker quick‑start, provides a web UI, REST API, SDKs, and production‑grade features (multi‑tenant auth, encrypted provider keys, safe migrations). Benchmarks show accuracy comparable to LightRAG but much faster cached queries.

  • Project
    Future-Scholars/paperlib

    An open-source academic paper management tool that simplifies metadata scraping and organization, featuring LLM-powered summarization and semantic search.

  • Project
    MarkPDFdown/markpdfdown

    A tool that uses multimodal LLMs to transcribe PDF files and images into well-formatted Markdown, preserving complex structures like tables and formulas.