[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.

LLM algorithm practice lab with theory, solutions, and test cases.《大模型算法与系统教程》面向大模型入门到进阶的算法实战教程，覆盖原理讲解、答案解析、测试用例与 CUDA/Triton 实战。

🌱 A tiny, readable LLM serving engine with vLLM/SGLang-style features.

Deterministic intermediate representation for AI agents — compile, verify, execute, and replay structured intent.

A lightweight Bun + Express template that connects to the Testune AI API and streams chat responses in real time using Server-Sent Events (SSE)

Ollama for one-GPU-pool boxes: unified-memory-safe admission control + single-resident hot-swap for local LLM serving on DGX Spark / GB10 / Jetson.

AI gateway and observability infrastructure for production LLM systems