Popular GitHub Repositories for Kernels

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

Efficient Triton Kernels for LLM Training

A collection of important graph embedding, classification and representation learning papers with implementations.

Deep learning in Rust, with shape checked tensors and neural networks

LLM notes, including model inference, transformer model structure, and llm framework code analysis notes.

NLP Zero to Hero in just 10 Kernels

This project is dedicated to the implementation and research of Kolmogorov-Arnold convolutional networks. The repository includes implementations of 1D, 2D, and 3D convolutions with different kernels, ResNet-like and DenseNet-like models, training code based on accelerate/PyTorch, as well as scripts for experiments with CIFAR-10 and Tiny ImageNet.

KernelBench: Can LLMs Write GPU Kernels? - Benchmark with Torch -> CUDA problems

[ICLR 2023] "More ConvNets in the 2020s: Scaling up Kernels Beyond 51x51 using Sparsity"; [ICML 2023] "Are Large Kernels Better Teachers than Transformers for ConvNets?"

Keras + Gaussian Processes: Learning scalable deep and recurrent kernels.

A simple NLP library allows profiling datasets with one or more text columns. When given a dataset and a column name containing text data, NLP Profiler will return either high-level insights or low-level/granular statistical information about the text in that column.

Learning kernels to maximize the power of MMD tests

LargeKernel3D: Scaling up Kernels in 3D Sparse CNNs (CVPR 2023)

Flash Dynamic Mask Attention

A light llama-like llm inference framework based on the triton kernel.

Spiking Neural Networks in C++ with strong GPU acceleration through CUDA

Attention Kernels for Symmetric Power Transformers

Open source cross-platform compiler for compute-intensive loops used in AI algorithms, from Microsoft Research

[CVPR 2023] LinK: Linear Kernel for LiDAR-based 3D Perception

Accelerate Multimodal AI | 3.7x faster E2E video gen and 1.6x faster E2E image gen | Custom sparse CUDA kernels 9.3x faster than FlashAttention-3 and 2.5x faster than cuBLAS