A reading list that keeps pace with the LLM systems arms race

What it does

This repository is a curated bibliography of academic papers, technical reports, and frameworks focused on the systems side of large language models — think training clusters with 10,000+ GPUs, inference serving stacks, and the scheduling glue that holds them together. It is organized by topic (training, serving, multimodal, RLHF) and by year, with direct links to PDFs and conference venues.

The interesting bit

The list is aggressively current: it already includes papers accepted to EuroSys and MLSys 2026, plus industrial technical reports from ByteDance, Meta, and Ant Group that often precede formal publication. The maintainer appears to track SOSP, OSDI, and MLSys program committees in real time.

Key highlights

• Coverage spans pre-training, post-training / RLHF, serving, agent systems, edge deployment, and multimodal systems
• Industrial lab reports sit alongside peer-reviewed work — useful for spotting production trends before they hit proceedings
• Includes frameworks (Megatron-LM, vLLM, SGLang) and benchmarks, not just research papers
• Sections on “LLM for Systems” and ML systems courses provide context for researchers crossing over
• Conference-specific tracking (e.g., NeurIPS 2025) with accepted papers listed before full publication

Caveats

• No annotation or synthesis — this is purely a link index, not a survey or reading guide
• Some entries are bare titles with venue tags; quality and relevance vary
• Industrial reports are included without editorial filtering beyond the maintainer’s judgment

Verdict

Worth bookmarking if you are a systems researcher, infrastructure engineer, or graduate student trying to stay current. Skip it if you want critical analysis or beginner-friendly explanations — this is a fire hose, not a filter.