Awesome LLM Systems Papers
A curated, continuously updated list of academic papers on large language model systems, tracking work on training, serving, agent systems and hardware co-design from 2024 through 2026.
Awesome LLM Systems Papers is a curated bibliography of academic papers, articles, tutorials and slides on the systems side of large language models: training (pre-training, post-training, fault tolerance), serving (including serving at the edge and for agent systems), multi-modal training and serving, LLM-for-systems work, industrial technical reports, ML conference proceedings, frameworks, benchmarks and leaderboards.
A trends section tracks how the field's attention shifted from 2024 to 2026 by counting papers per technique and year: serving remained the largest area by absolute count but its share fell from 49% to 33%, while kernel and model co-design, agentic systems, AI-for-systems and edge serving grew fastest, and agentic and multi-agent techniques rose from about one percent to thirteen percent of the year's papers.
The list is organized by table of contents into training, serving, systems and reading sections, and reproduction scripts for the trend analysis are included in the repository.