Lists (3)
Stars
Disaggregated Prefill-Decode LLM serving engine on mini-sglang with a proved KV page state machine (PdLedger)
斯坦福 Stanford CS329Z Engineering AI Agents 中文教程：面向 Python 开发者的 AI 智能体课程，19 讲与 4 项实践，涵盖 LLM、RAG、工具调用、记忆、多智能体、评测与安全。完整 Markdown、在线与离线 HTML。Chinese course notes.
An experimental implementation of compiler-driven automatic sharding of models across a given device mesh.
An Open Compiler Harness for Agentic GPU Programming
CUA-Master: Scaling Computer-use Agents through Open-World Recursive Improvements
SoL-Pi: Scaling Auto-Research Loops for Efficient Agent Harnesses
SpecPilot: Autonomous Algorithm Discovery and Systems Optimization for Speculative Decoding
A framework for building robust RL environments to train aligned AI.
Towards a More Autonomous Agent System
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
DeepInfer: High-performance, large-scale inference framework for DeepSeek models
AI reads books: Page-by-Page PDF Knowledge Extractor & Summarizer. script performs an intelligent page-by-page analysis of PDF books, methodically extracting knowledge points and generating progres…
Automatic formal verification for warp-specialized GPU kernels.
Dist-MoE implements Speed-of-Light distributed Mixture-of-Experts forward and backward kernels. They are highly optimized and fused comm-compute kernels for training and inference on SM100+ GPUs
Lightweight, agent-native sandboxing with diff, commit and abort
The Little Book of Reinforcement Learning
A sparse-first inference engine (SparseEngine).
Specula: An agentic tool for finding deep bugs in system code using TLA+
Reverse engineer anything with agents, from app behavior down to native binaries.
Turn long-horizon goals into verifiable results. A local-first semantic control plane and research workbench for long-running agent teams, across sessions and runtimes.
GraphTok: exact BPE tokenization on the GPU as one replayable CUDA graph, with vLLM and NVIDIA Dynamo integrations