Understanding R1-Zero-Like Training: A Critical Perspective

PyTorch-native post-training at scale

LeetGPU Challenges

A Survey of Efficient Attention Methods: Hardware-efficient, Sparse, Compact, and Linear Attention

Speed Always Wins: A Survey on Efficient Architectures for Large Language Models

Triton based sparse quantization attention kernel collection

Distributed parallel 3D-Causal-VAE for efficient training and inference

High performance inference engine for diffusion models

Toolchain built around the Megatron-LM for Distributed Training

Rust crates for XetHub

A curated list of recent papers on efficient video attention for video diffusion models, including sparsification, quantization, and caching, etc.

Paper list in the survey: A Survey on Vision-Language-Action Models: An Action Tokenization Perspective

Utility scripts for PyTorch (e.g. Make Perfetto show some disappearing kernels, Memory profiler that understands more low-level allocations such as NCCL, ...)

Omni_Infer is a suite of inference accelerators designed for the Ascend NPU platform, offering native support and an expanding feature set.

Making Flux go brrr on GPUs.

Fastest kernels written from scratch

slime is an LLM post-training framework for RL Scaling.

Puzzles for learning Triton

Yinghan's Code Sample

CUDA Matrix Multiplication Optimization

Fast CUDA matrix multiplication from scratch

Easier, quicker command-line CUDA profiling

Bridge Megatron-Core to Hugging Face/Reinforcement Learning

Nano vLLM

A course of learning LLM inference serving on Apple Silicon for systems engineers: build a tiny vLLM + Qwen.

Graphic notes on Gilbert Strang's "Linear Algebra for Everyone"