LMDeploy is a toolkit for compressing, deploying, and serving LLMs.

Kernel Tuner

Triton implementation of FlashAttention2 that adds Custom Masks.

Implementation of ConjugateGradients method using C and Nvidia CUDA

Faster Pytorch bitsandbytes 4bit fp4 nn.Linear ops

This repository contains examples CUDA usage in Cython code.

This is a cross-chip platform collection of operators and a unified neural network library.

Code repository for ICLR 2025 paper "LeanQuant: Accurate and Scalable Large Language Model Quantization with Loss-error-aware Grid"

Pytorch implementation of a message passing neural network with RNN sub-units

Object Tracking using GPU acceleration.

This project integrates a custom CUDA-based matrix multiplication kernel into a PyTorch deep learning model, leveraging GPU acceleration for matrix operations. The goal is to compare the performance of this custom kernel with PyTorch's built-in matrix multiplication and demonstrate how custom CUDA kernels can optimize compute-intensive operations.

Spiral's Machine Learning Library

KernelHeim – development ground of custom Triton and CUDA kernel functions designed to optimize and accelerate machine learning workloads on NVIDIA GPUs. Inspired by the mythical stronghold of the gods, KernelHeim is a forge where high-performance kernels are crafted to unlock the full potential of the hardware.

SParry is a shortest path calculating Python tool using some algorithms with CUDA to speedup.

Fused 4bit AdamW in Cuda

GPU porgamming CUDA is the repo that has all the list of my materials that I used for the CUDA . I learned CUDA myself and this material helped me get the basic strong .

CNNs for spectrogram-based music recommendation (Undergraduate dissertation)

The provided code is a Python script that uses the CuPy library to perform optimized GPU operations, specifically matrix multiplication. The script includes a custom CUDA kernel that is optimized for performance and energy consumption. The kernel uses half-precision floating-point numbers (float16) for improved performance and warp utilization.

High-performance 2D Quantum Dot (QD) Simulator implemented in C++ and Python

A collection of ultra-simple yet high-performance CUDA kernels.