This repository is a curated collection of research papers focused on machine learning systems, targeting researchers, engineers, and practitioners in the field. It provides a structured overview of advancements in areas like distributed training, inference optimization, and specialized ML workloads, serving as a valuable resource for staying abreast of system-level innovations.

How It Works

The repository organizes papers by key ML system categories, including data processing, training systems, inference systems, and specialized architectures like Mixture of Experts (MoE) and Long Context LLMs. Each entry typically includes the paper's title, publication venue (e.g., OSDI, SIGMOD, NeurIPS), and year, allowing users to quickly identify relevant research.

Quick Start & Requirements

• Access: No installation required. Browse papers directly via the GitHub repository.
• Requirements: A web browser and internet connection.
• Links: The README includes links to related curated lists and resources for further exploration.

Highlighted Details

• Comprehensive coverage of ML systems research, spanning over 13 major categories.
• Detailed sub-categorization within each major topic (e.g., DNN job scheduling, GPU sharing, communication optimization).
• Extensive listing of papers from top-tier conferences and journals in systems and machine learning.
• Includes links to related "awesome" lists for broader context.

Maintenance & Community

The repository is a static collection of links, with its primary motivation stemming from several existing "awesome" lists in ML systems and related fields. There is no explicit mention of active maintenance or community interaction channels.

Licensing & Compatibility

The repository itself, as a collection of links to external papers, does not have a specific license. The linked papers are subject to their respective publisher's copyright and licensing terms.

Limitations & Caveats

This repository is a curated list of papers and does not provide code, implementations, or direct access to the paper content itself. Users must follow the links to access the full papers, which may be behind paywalls or require institutional access.