Microbenchmarking hyperparameter tuning for JAX functions.

GPUGrants - a list of GPU grants that I can think of

DoubleAI’s hyperoptimised version of cuGraph

Automated High-Performance GPU Kernel Generation

SGLang is a high-performance serving framework for large language models and multimodal models.

A Quirky Assortment of CuTe Kernels

UCCL is an efficient communication library for GPUs, covering collectives, P2P (e.g., KV cache transfer, RL weight transfer), and EP (e.g., GPU-driven)

A PyTorch-native inference engine with cache, parallelism, quantization and cpu offload for DiTs.

DFlash: Block Diffusion for Flash Speculative Decoding

Artifact for "Fail Fast, Win Big: Rethinking the Drafting Strategy in Speculative Decoding via Diffusion LLMs" [arXiv '25]

Codebase for Cuda Learning

A curriculum for learning about gpu performance engineering, from scratch to what the frontier AI labs do

how to optimize some algorithm in cuda.

System Intelligence Benchmark

Efficient Long-context Language Model Training by Core Attention Disaggregation

A compact implementation of SGLang, designed to demystify the complexities of modern LLM serving systems.

Miles is an enterprise-facing reinforcement learning framework for LLM and VLM post-training, forked from and co-evolving with slime.

Nano vLLM

Mirage Persistent Kernel: Compiling LLMs into a MegaKernel

Accelerate inference without tears

TPU inference for vLLM, with unified JAX and PyTorch support.

NanoGPT (124M) in 90 seconds

Tenstorrent MLIR compiler

🤘 TT-NN operator library, and TT-Metalium low level kernel programming model.

Universal LLM Deployment Engine with ML Compilation

Open Machine Learning Compiler Framework