Aibek Bekbayev

MoonEP. MoonEP: A Perfectly Balanced Expert Parallelism Library via Dynamic Redundant Experts

low-latency-nccl. C++

gigatoken. Language model tokenization at GB/s

lingbot-vision. Self-supervised learning for spatial perception

tml-fa4. FA4-based Relative Attention Kernel developed by TML and Colfax

ShapeR. Code for the ShapeR research paper

swav. PyTorch implementation of SwAV https//arxiv.org/abs/2006.09882

krea-2. Python

lingbot-video. Scaling Mixture-of-Experts Video Pretraining for Embodied Intelligence

SimCLR. PyTorch implementation of SimCLR: A Simple Framework for Contrastive Learning of Visual Representations

dinov3. Reference PyTorch implementation and models for DINOv3

UnifiedReward. Official implementation of UnifiedReward & [NeurIPS 2025] UnifiedReward-Think & UnifiedReward-Flex

DanceGRPO. An official implementation of DanceGRPO: Unleashing GRPO on Visual Generation

Helios. Helios: Real Real-Time Long Video Generation Model

gdrcopy. A fast GPU memory copy library based on NVIDIA GPUDirect RDMA technology

Megakernels. Kernels, of the mega variety :)

Etha. Python

PipelineRL. A scalable asynchronous reinforcement learning implementation with in-flight weight updates.

hermes-agent. The agent that grows with you

fastgrpo. Python

lucebox. LLM speculative inference server for consumer hardware & heterogeneous computing

gemlite. Fast low-bit matmul kernels in Triton

marlin. FP16xINT4 LLM inference kernel that can achieve near-ideal ~4x speedups up to medium batchsizes of 16-32 tokens.

FlashQLA. high-performance linear attention kernel library built on TileLang

flashinfer. FlashInfer: Kernel Library for LLM Serving

OpenClaw-RL. OpenClaw-RL: Train any agent simply by talking

ares. ares is a cross-platform, open source, multi-system emulator, focusing on accuracy and preservation.

tokenspeed. TokenSpeed is a speed-of-light LLM inference engine.

SEAL. Self-Adapting Language Models

context-1-data-gen. Python

Search-R1. Search-R1: An Efficient, Scalable RL Training Framework for Reasoning & Search Engine Calling interleaved LLM based on veRL

SZ3. Error-bounded Lossy Data Compressor (for floating-point/integer datasets)

cuSZ. A GPU accelerated error-bounded lossy compression for scientific data.

flash-attention-residuals. Triton kernels and PyTorch ops for Block Attention Residuals (AttnRes)

cuSZp. Fast GPU error-bounded lossy compressor for floating-point data.

PLFM_RADAR. Open-source, low-cost 10.5 GHz PLFM phased array RADAR system

kraken. Triton-based Symmetric Memory operators and examples

xccl. C

DualPipe. A bidirectional pipeline parallelism algorithm for computation-communication overlap in DeepSeek V3/R1 training.

DeepEP. DeepEP: an efficient expert-parallel communication library

firecracker. Secure and fast microVMs for serverless computing.

prime-rl. Agentic RL Training at Scale

libTAS. GNU/Linux software to (hopefully) give TAS tools to games

llm-code-editor. LLM Code Editor (think Claude Code)

KazakhstanOlympiadAI-HomeTask. Python

transitions. A lightweight, object-oriented finite state machine implementation in Python with many extensions

ml_conformer_generator. A tool for spatially-aware molecule design and optimisation via Equivariant Diffusion and GCN

flightmare. An Open Flexible Quadrotor Simulator

Colosseum. Open source simulator for autonomous robotics built on Unreal Engine with support for Unity

freat. A simple game hacking tool, leveraging Godot and Frida for cross-platform support and :star: style points :star:.

pysc2. StarCraft II Learning Environment

PhoenixOS. Fast OS-level support for GPU checkpoint and restore

selkies. Open-Source Low-Latency Accelerated Linux WebRTC HTML5 Remote Desktop Streaming Platform for Self-Hosting, Containers, Kubernetes, or Cloud/HPC

qemu. QEMU in a Docker container.

Gymnasium. A standard API for single-agent reinforcement learning environments, with popular reference environments and related utilities (formerly Gym)

shardy. MLIR-based partitioning system

einops. Flexible and powerful tensor operations for readable and reliable code (for pytorch, jax, TF and others)

scalable-dreamer. Python

LTX-2. Official Python inference and LoRA trainer package for the LTX-2 audio–video generative model.

triton-ext. A collection of out-of-tree extensions for the Triton language and compiler

torchtitan. A PyTorch native platform for training generative AI models

comfy-kitchen. Fast kernel library for Diffusion inference with multiple compute backends.

cutile-python. cuTile is a programming model for writing parallel kernels for NVIDIA GPUs

DistCA. Efficient Long-context Language Model Training by Core Attention Disaggregation

sonic-moe. Accelerating MoE with IO and Tile-aware Optimizations

Duix-Mobile. 🚀 The best real-time interactive AI avatar(digital human) with on-premise deployment and <1.5 s latency.

rocSHMEM. [DEPRECATED] Moved to ROCm/rocm-systems repo

clice. A next-generation C++ language server for modern C++, focused on high performance and deep code intelligence

nvbench. CUDA Kernel Benchmarking Library

anny. Anny, A Free and Interpretable Human Body Model for all ages, written in PyTorch.

HLA. Official Project Page for HLA: Higher-order Linear Attention (https://arxiv.org/abs/2510.27258)

flash-linear-attention. 🚀 Efficient implementations for emerging model architectures

tritonparse. TritonParse: A Compiler Tracer, Visualizer, and Reproducer for Triton Kernels

botorch. Bayesian optimization in PyTorch

Code2Video. [ICML 2026] Video generation via code

multi-gpu-programming-models. Examples demonstrating available options to program multiple GPUs in a single node or a cluster

llmq. Quantized LLM training in pure CUDA/C++.

ffpa-attn. FFPA: Kernel Library for Large Headdim Attention - 1.5x~6x speedup over PyTorch SDPA.

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

triton. Github mirror of trition-lang/triton repo.

nvshmem. NVIDIA NVSHMEM is a parallel programming interface for NVIDIA GPUs based on OpenSHMEM. NVSHMEM can significantly reduce multi-process communication and coordination overheads by allowing programmers to perform one-sided communication from within CUDA kernels and on CUDA streams.

accelerated-computing-hub. NVIDIA curated collection of educational resources related to general purpose GPU programming.

flux. A fast communication-overlapping library for tensor/expert parallelism on GPUs.

mirage. Mirage Persistent Kernel: Compiling LLMs into a MegaKernel

FPGA-Project-2022-simple-tpu. Systolic array based simple TPU for CNN on PYNQ-Z2

SimpleTPU. A FPGA Based CNN accelerator, following Google's TPU V1.

Triton-distributed. Distributed Compiler based on Triton for Parallel Systems

real-work. Real Work is a framework for RL environments that simulate real work.

cccl. CUDA Core Compute Libraries

LTX-Video. Official repository for LTX-Video

ring-flash-attention. Ring attention implementation with flash attention

pplx-kernels. Perplexity GPU Kernels

attention-gym. Helpful tools and examples for working with flex-attention

efficientvit. Efficient vision foundation models for high-resolution generation and perception.

tuning_playbook. A playbook for systematically maximizing the performance of deep learning models.

torch-dct. DCT (discrete cosine transform) functions for pytorch

sycl-tla. SYCL* Templates for Linear Algebra (SYCL*TLA) - SYCL based CUTLASS implementation for Intel GPUs

wmma_tensorcore_sample. Matrix Multiply-Accumulate with CUDA and WMMA( Tensor Core)

accelerated-model-architectures. Python

cuda_hgemm. Several optimization methods of half-precision general matrix multiplication (HGEMM) using tensor core with WMMA API and MMA PTX instruction.

flow_grpo. [NeurIPS 2025] An official implementation of Flow-GRPO: Training Flow Matching Models via Online RL

tpux. A set of Python scripts that makes your experience on TPU better

MagiAttention. A Distributed Attention Towards Linear Scalability for Ultra-Long Context, Heterogeneous Data Training

ComfyUI_ProfilerX. Node and workflow profiling. Find bottlenecks in your workflows. See trends over time.

LTX-Video-Trainer. Community trainer for Lightricks' LTX Video model 🎬 ⚡️

LTX-Video-Q8-Kernels. Python

FasterTransformer. Transformer related optimization, including BERT, GPT

ComfyUI-LTXVideo. LTX-Video Support for ComfyUI

DeepGEMM. DeepGEMM: clean and efficient BLAS kernel library on GPU

yamajiflow. Python

PolyDye. Full Color Printer Mod for Marlin 3D Printers

Hunyuan3D-2. High-Resolution 3D Assets Generation with Large Scale Hunyuan3D Diffusion Models.

SPAN. This repository is an official PyTorch implementation of our paper “Lightweight Image Super-Resolution with Sliding Proxy Attention Network”

flow_matching. A PyTorch library for implementing flow matching algorithms, featuring continuous and discrete flow matching implementations. It includes practical examples for both text and image modalities.

unsloth. Unsloth is a local UI for training and running Kimi K3, Gemma 4, Qwen3.6, DeepSeek-V4, GLM and other models.

FastSoftmax. Step by step implementation of a fast softmax kernel in CUDA

terrain-navigation. Implementation for safe low altitude navigation in steep terrain for fixed-wing Aerial Vehicles

LTX-Video. LTXVideo Q8

q8_kernels. Cuda

VideoSys. VideoSys: An easy and efficient system for video generation

channels-last-groupnorm. A CUDA kernel for NHWC GroupNorm for PyTorch

Parallel-Computing-Cuda-C. CUDA Learning guide

CogVideo. text and image to video generation: CogVideoX (2024) and CogVideo (ICLR 2023)

triton. Development repository for the Triton language and compiler

stable-fast. https://wavespeed.ai/ Best inference performance optimization framework for HuggingFace Diffusers on NVIDIA GPUs.

MambaVision. [CVPR 2025] Official PyTorch Implementation of MambaVision: A Hybrid Mamba-Transformer Vision Backbone

NATTEN. Fast Multi-dimensional Sparse Attention

FALCONN. FAst Lookups of Cosine and Other Nearest Neighbors (based on fast locality-sensitive hashing)

mra-attention. Python

cuda-checkpoint. CUDA checkpoint and restore utility

Open-Sora-Plan. This project aim to reproduce Sora (Open AI T2V model), we wish the open source community contribute to this project.

diffae. Official implementation of Diffusion Autoencoders

pytorch-binary-converter. Turning float tensors to binary tensors according to IEEE-754 standard.

jepa. PyTorch code and models for V-JEPA self-supervised learning from video.

phenaki. A phenaki reproduction using pytorch.

co-tracker. CoTracker is a model for tracking any point (pixel) on a video.

AnimateDiff. Official implementation of AnimateDiff.

common_metrics_on_video_quality. You can easily calculate FVD, PSNR, SSIM, LPIPS for evaluating the quality of generated or predicted videos.

Latte. [TMLR 2025] Latte: Latent Diffusion Transformer for Video Generation.

Latte. The official implementation of Latte: Latent Diffusion Transformer for Video Generation.

tarproc. Utilities for sequential processing of tar files.

webdataset. A high-performance Python-based I/O system for large (and small) deep learning problems, with strong support for PyTorch.

Tensor-Puzzles. Solve puzzles. Improve your pytorch.

agile_autonomy. Repository Containing the Code associated with the Paper: "Learning High-Speed Flight in the Wild"

QuestPatcher. Generic il2cpp modding tool for Oculus Quest (1/2/3) apps.

LucidDreamer. Official code for the paper "LucidDreamer: Domain-free Generation of 3D Gaussian Splatting Scenes".

animatediff-cli-prompt-travel. animatediff prompt travel

PASD. [ECCV2024] Pixel-Aware Stable Diffusion for Realistic Image Super-Resolution and Personalized Stylization

cutlass. CUDA Templates and Python DSLs for High-Performance Linear Algebra

higgsfield. Fault-tolerant, highly scalable GPU orchestration, and a machine learning framework designed for training models with billions to trillions of parameters

IP-Adapter. The image prompt adapter is designed to enable a pretrained text-to-image diffusion model to generate images with image prompt.

hack-assembler. Compiler for simple assembly language

LLaMPPL. A domain-specific probabilistic programming language for modeling and inference with language models

awesome-image-translation. Collection of awesome resources on image-to-image translation.

HS-Diffusion. HS-Diffusion: Semantic-Mixing Diffusion for Head Swapping

dcface. Python

StyleMask. Authors official PyTorch implementation of the "StyleMask: Disentangling the Style Space of StyleGAN2 for Neural Face Reenactment" [FG 2023].

manifesto. The OpenTF Manifesto expresses concern over HashiCorp's switch of the Terraform license from open-source to the Business Source License (BSL) and calls for the tool's return to a truly open-source license.

latent-pose-reenactment. The authors' implementation of the "Neural Head Reenactment with Latent Pose Descriptors" (CVPR 2020) paper.

e4s. (CVPR 2023) E4S: Fine-grained Face Swapping via Regional GAN Inversion

alga. Algebraic graphs

facechain. FaceChain is a deep-learning toolchain for generating your Digital-Twin.

HyperKohaku. A diffusers based implementation of HyperDreamBooth

perfusion-pytorch. Implementation of Key-Locked Rank One Editing, from Nvidia AI

flash-attention. Fast and memory-efficient exact attention

sd-scripts. Python

RecurrentGPT. Official Code for Paper: RecurrentGPT: Interactive Generation of (Arbitrarily) Long Text

Instruction-Tuning-Papers. Reading list of Instruction-tuning. A trend starts from Natrural-Instruction (ACL 2022), FLAN (ICLR 2022) and T0 (ICLR 2022).

Awesome-Story-Generation. This repository collects an extensive list of awesome papers about Story Generation / Storytelling, exclusively focusing on the era of Large Language Models (LLMs).

Awesome-Controllable-T2I-Diffusion-Models. A collection of resources on controllable generation with text-to-image diffusion models.

awesome-personalization. ✨ A curated list of awesome things related to personalization.

prompt-in-context-learning. Awesome resources for in-context learning and prompt engineering: Mastery of the LLMs such as ChatGPT, GPT-3, and FlanT5, with up-to-date and cutting-edge updates.

RETRO-pytorch. Implementation of RETRO, Deepmind's Retrieval based Attention net, in Pytorch

tree-of-thought-llm. [NeurIPS 2023] Tree of Thoughts: Deliberate Problem Solving with Large Language Models

audiocraft. Audiocraft is a library for audio processing and generation with deep learning. It features the state-of-the-art EnCodec audio compressor / tokenizer, along with MusicGen, a simple and controllable music generation LM with textual and melodic conditioning.

trl. Train transformer language models with reinforcement learning.

ViCo. Official PyTorch codes for the paper: "ViCo: Detail-Preserving Visual Condition for Personalized Text-to-Image Generation"

llm-numbers. Numbers every LLM developer should know

fastcomposer. [IJCV] FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attention

Stable-Diffusion-Inpaint. Stable diffusion for inpainting

Awesome-Transformer-Attention. An ultimately comprehensive paper list of Vision Transformer/Attention, including papers, codes, and related websites

tomesd. Speed up Stable Diffusion with this one simple trick!

dinov2. PyTorch code and models for the DINOv2 self-supervised learning method.

LLM-As-Chatbot. LLM as a Chatbot Service

PAIR-Diffusion. [CVPR 2024] PAIR Diffusion: A Comprehensive Multimodal Object-Level Image Editor

multimodal-garment-designer. This is the official repository for the paper "Multimodal Garment Designer: Human-Centric Latent Diffusion Models for Fashion Image Editing". ICCV 2023

TurkicASR. A multilingual ASR model that can recognize ten Turkic languages—Azerbaijani, Bashkir, Chuvash, Kazakh, Kyrgyz, Sakha, Tatar, Turkish, Uyghur, and Uzbek.

FixNoise. Official Pytorch Implementation for "Fix the Noise: Disentangling Source Feature for Controllable Domain Translation" (CVPR 2023, CVPRW 2022 Best paper)

DreamArtist-stable-diffusion. stable diffusion webui with contrastive prompt tuning

Text2Poster-ICASSP-22. Official implementation of the ICASSP-2022 paper "Text2Poster: Laying Out Stylized Texts on Retrieved Images"

sd_personalization_encoder. Unofficial implementation of Encoder-based Domain Tuning for Fast Personalization of Text-to-Image Models

ELITE. ELITE: Encoding Visual Concepts into Textual Embeddings for Customized Text-to-Image Generation (ICCV 2023, Oral)

TaskMatrix. Python

GLIGEN. Open-Set Grounded Text-to-Image Generation

SSHarmonization. [ICCV'2021] "SSH: A Self-Supervised Framework for Image Harmonization", Yifan Jiang, He Zhang, Jianming Zhang, Yilin Wang, Zhe Lin, Kalyan Sunkavalli, Simon Chen, Sohrab Amirghodsi, Sarah Kong, Zhangyang Wang

DeepImageBlending. This is a Pytorch implementation of deep image blending

sparsednn. Fast sparse deep learning on CPUs

ProDiff. PyTorch Implementation of ProDiff (ACM-MM'22) with a Extremely-Fast diffusion speech synthesis pipeline

Empathy-Mental-Health. Repository containing codes and dataset access instructions for the EMNLP 2020 paper on empathy in text-based mental health support

Paella. Official Implementation of Paella https://arxiv.org/abs/2211.07292v2