Pinned Repositories
intel-extension-for-transformers
⚡ Build your chatbot within minutes on your favorite device; offer SOTA compression techniques for LLMs; run LLMs efficiently on Intel Platforms⚡
cutlass
CUDA Templates for Linear Algebra Subroutines
FastAPSP
The Fast APSP algorithm is used to solve the All-Pairs Shortest Paths (APSP) problem. The algorithm uses the divide and conquers strategy. First, divide the graph structure by METIS, and divide the input graph G into multiple subgraphs. Then the solution of the APSP problem is solved by computing the subgraph. The Fast APSP algorithm combines the SSSP algorithm and the Floyd-Warshall algorithm. Compared with the Part APSP algorithm, it eliminates the data dependence and communication between sub-graphs. The Fast APSP algorithm has achieved good performance in graphs with good properties. We tested a lot of sparse graph data in the Suite sparse matrix collection and network repository, and the Fast APSP algorithm showed better performance than other APSP algorithms.
How_to_optimize_in_GPU
This is a series of GPU optimization topics. Here we will introduce how to optimize the CUDA kernel in detail. I will introduce several basic kernel optimizations, including: elementwise, reduce, sgemv, sgemm, etc. The performance of these kernels is basically at or near the theoretical limit.
OpenBLAS
OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.
Paddle
PArallel Distributed Deep LEarning: Machine Learning Framework from Industrial Practice （『飞桨』核心框架，深度学习&机器学习高性能单机、分布式训练和跨平台部署）
pytorch
Tensors and Dynamic neural networks in Python with strong GPU acceleration
tvm
Open deep learning compiler stack for cpu, gpu and specialized accelerators
deepxde
A library for scientific machine learning and physics-informed learning
Paddle
PArallel Distributed Deep LEarning: Machine Learning Framework from Industrial Practice （『飞桨』核心框架，深度学习&机器学习高性能单机、分布式训练和跨平台部署）
Liu-xiandong's Repositories
Liu-xiandong/How_to_optimize_in_GPU
This is a series of GPU optimization topics. Here we will introduce how to optimize the CUDA kernel in detail. I will introduce several basic kernel optimizations, including: elementwise, reduce, sgemv, sgemm, etc. The performance of these kernels is basically at or near the theoretical limit.
Liu-xiandong/FastAPSP
The Fast APSP algorithm is used to solve the All-Pairs Shortest Paths (APSP) problem. The algorithm uses the divide and conquers strategy. First, divide the graph structure by METIS, and divide the input graph G into multiple subgraphs. Then the solution of the APSP problem is solved by computing the subgraph. The Fast APSP algorithm combines the SSSP algorithm and the Floyd-Warshall algorithm. Compared with the Part APSP algorithm, it eliminates the data dependence and communication between sub-graphs. The Fast APSP algorithm has achieved good performance in graphs with good properties. We tested a lot of sparse graph data in the Suite sparse matrix collection and network repository, and the Fast APSP algorithm showed better performance than other APSP algorithms.
Liu-xiandong/cutlass
CUDA Templates for Linear Algebra Subroutines
Liu-xiandong/OpenBLAS
OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.
Liu-xiandong/Paddle
PArallel Distributed Deep LEarning: Machine Learning Framework from Industrial Practice （『飞桨』核心框架，深度学习&机器学习高性能单机、分布式训练和跨平台部署）
Liu-xiandong/pytorch
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Liu-xiandong/tvm
Open deep learning compiler stack for cpu, gpu and specialized accelerators