mlsys

Machine Learning Systems: Foundations, Scaling, Agentic AI, and Physical AI (Vols I–IV) • Harvard CS249r | https://mlsysbook.ai

AISystem 主要是指AI系统，包括AI芯片、AI编译器、AI推理和训练框架等AI全栈底层技术

The RL Bridge for LLM-based Agent Applications. Made Simple & Flexible.

🚀 Awesome System for Machine Learning ⚡️ AI System Papers and Industry Practice. ⚡️ System for Machine Learning, LLM (Large Language Model), GenAI (Generative AI). 🍻 OSDI, NSDI, SIGCOMM, SoCC, MLSys, etc. 🗃️ Llama3, Mistral, etc. 🧑‍💻 Video Tutorials.

[ICLR2025 Spotlight] SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

ComfyUI Plugin of Nunchaku

[ICML2025] SpargeAttention: A training-free sparse attention that accelerates any model inference.

A model compilation solution for various hardware

SLA: Beyond Sparsity in Diffusion Transformers via Fine-Tunable Sparse–Linear Attention

FedScale is a scalable and extensible open-source federated learning (FL) platform.

[ACL 2026] Towards Efficient Large Language Model Serving: A Survey on System-Aware KV Cache Optimization

Measure and optimize the energy consumption of your AI applications!

The repository has collected a batch of noteworthy MLSys bloggers (Algorithms/Systems)

A ChatGPT(GPT-3.5) & GPT-4 Workload Trace to Optimize LLM Serving Systems

Machine Learning Framework for Operating Systems - Brings ML to Linux kernel

An acceleration library that supports arbitrary bit-width combinatorial quantization operations

A scalable & efficient active learning/data selection system for everyone.

An agent harness that compiles a model into one provably-correct, self-retargeting CUDA megakernel and self-tunes it past cuBLAS at batch-1 LLM decode, paper: https://arxiv.org/abs/2606.09682

Code for Paper "RoboECC: Multi-Factor-Aware Edge-Cloud Collaborative Deployment for VLA Models" accepted by IJCNN 2026.