zhang

hello-agents. 📚 《从零开始构建智能体》——从零开始的智能体原理与实践教程

skills. Skills for Real Engineers. Straight from my .claude directory.

SimuMax. a static analytical model for LLM distributed training

evergreen-skills-developers. List of evergreen skills, based on software development best practices & cross-framework principles, that should serve as a fair assessment of skilled software engineers / developers

CUDA-Learn-Notes. 📚200+ Tensor/CUDA Cores Kernels, ⚡️flash-attn-mma, ⚡️hgemm with WMMA, MMA and CuTe (98%~100% TFLOPS of cuBLAS/FA2 🎉🎉).

DiffSynth-Engine. Python

xllm. A high-performance inference engine for LLM, VLM, DiT and REC models, optimized for diverse AI accelerators. It is hosted in OpenAtom Foundation.

skills. Public repository for Agent Skills

playground. playground tasks for newcomers

VibeKernel. Benchmarking coding-agent paradigms (plain prompt, /goal, skills, autoresearch...) on one fixed task: hand-writing the fastest fp16 GEMM kernel for the NVIDIA A100.

deepseek-v4-deep-dive. DeepSeek V4 深度解读 · 73 页 PPT + 20 分钟讲稿 + 发布动画 · Designed with huashu-design

OriginDL. Implement a Pytorch-like DL library in C++ from scratch, step by step

awesome-claude-skills. A curated list of awesome Claude Skills, resources, and tools for customizing Claude AI workflows

claude-mem. Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More

torchtitan. A PyTorch native platform for training generative AI models

accel-sim-framework. This is the top-level repository for the Accel-Sim framework.

LPLB. An early research stage expert-parallel load balancer for MoE models based on linear programming.

LLMCompass. Python

autoclip. AutoClip : AI-powered video clipping and highlight generation · 一款智能高光提取与剪辑的二创工具

FunClip. FunASR-powered video transcription, subtitle generation, and LLM-assisted clipping tool with a local Gradio UI.

NarratoAI. 利用AI大模型，一键解说并剪辑视频； Using AI models to automatically provide commentary and edit videos with a single click.

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

easy-vibe. 💻 vibe coding 2026 | Your First Modern Coding course beginners to master step by step.

supervision. We write your reusable computer vision tools. 💜

mini-coding-agent. Minimal and readable coding agent harness implementation in Python to explain the core components of coding agents.

claw-code. An agent-managed museum exhibit, built in Rust with Gajae-Code / LazyCodex — developed and maintained with no human intervention.

vllm. A high-throughput and memory-efficient inference and serving engine for LLMs

AI-Infra-Auto-Driven-SKILLS. Python

how-claude-code-works. Deep dive into Claude Code internals — architecture, agent loop, context engineering, and more. / 深入解析 Claude Code 源码：架构、Agent 循环、上下文工程、工具系统等

MSA. Memory Sparse Attention - A scalable, end-to-end trainable latent-memory framework for 100M-token contexts.

autoresearch. AI agents running research on single-GPU nanochat training automatically

LLMs-from-scratch. Implement a ChatGPT-like LLM in PyTorch from scratch, step by step

torchvista. Interactive Pytorch forward pass visualization in notebooks

mfu_calculation. A simple calculation for LLM MFU.

InfraTech. 分享AI Infra知识&代码练习：PyTorch、vLLM/SGLang、slime/vime框架入门⚡️、性能加速🚀、大模型基础🧠、AI软硬件🔧等

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

TelegramGroup. 2026最新悄咪咪收集的10000+个Telegram群合集、tg导航，附全网最有趣好用的机器人BOT🤖【tg10000.com】

CUDATutorial. A self-learning tutorail for CUDA High Performance Programing.

openclaw. Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞

vllm-omni. A framework for efficient model inference with omni-modality models

hpc-ops. High Performance LLM Inference Operator Library

tilelang. Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels

nanochat. The best ChatGPT that $100 can buy.

DLPerf. DeepLearning Framework Performance Profiling Toolkit

apex_plus. APEX+ is an LLM Serving Simulator

vidur. Accurate, large-scale, and extensible simulator for LLM inference Systems

SCALE-Sim. Repository to host and maintain SCALE-Sim code

aiconfigurator. Offline optimization of your disaggregated Dynamo graph

DPCIManager. Simple OS X app for viewing PCI hardware info

pcileech-fpga. FPGA modules used together with the PCILeech Direct Memory Access (DMA) Attack Software

pcileech-wifi. pcileech-fpga with wireless card emulation

PCILeech-DMA-Firmware. The last Pcileech DMA CFW guide you will ever need. Sponsored by DMAPolice.com

nano-vllm. Nano vLLM

Cortex. 从零构建大模型：从预训练到RLHF的完整实践

build-your-claude-code-from-scratch. Build a Claude Code–like CLI coding agent from scratch.

reefy. Turn your PC into a Reef - run OpenClaw, Ollama, and more on your own PC. Private. Local. Open source. Simple.

picotron_tutorial. Python

picotron. Minimalistic 4D-parallelism distributed training framework for education purpose

YJYpaper. 一个用来记录武汉大学杨景媛论文问题的仓库

DeepSeek_Simulator. Python

quarkdown. 🪐 Markdown with superpowers: from ideas to papers, presentations, websites, books, and knowledge bases.

GPUs-Specs. Summary of the Specs of Commonly Used GPUs for Training and Inference of LLM

SimAI. Python

sgl-learning-materials. Materials for learning SGLang

DeepSeek-MoE. DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models

bitsandbytes. Accessible large language models via k-bit quantization for PyTorch.

shallowsim. DeepSeek-V3/R1 inference performance simulator

codex. Lightweight coding agent that runs in your terminal

lite_lang. 一个轻量化的大模型推理框架

AderXCoding. 介绍各类语言，库，系统编程以及算法的学习

torch_mlu. Python

torch_musa. torch_musa is an open source repository based on PyTorch, which can make full use of the super computing power of MooreThreads graphics cards.

pytorch. Tensors and Dynamic neural networks in Python with strong GPU acceleration

miniforge. A conda-forge distribution.

CThreadPool. 【A simple used C++ threadpool】一个简单好用，性能优异的，跨平台的C++线程池。欢迎 star & fork

Awesome-ML-SYS-Tutorial. My learning notes for ML SYS.

DeepEP. DeepEP: an efficient expert-parallel communication library

DeepSeek-V3. Python

LLMSpeculativeSampling. Fast inference from large lauguage models via speculative decoding

LLMs-Zero-to-Hero. 从无名小卒到大模型（LLM）大英雄~ 欢迎关注后续！！！

OptimizeDGEMM. C

deepseekv2-profile. Jupyter Notebook

CUDA_Kernel_Samples. CUDA 算子手撕与面试指南

LeetCUDA. 📚LeetCUDA: Modern CUDA Learn Notes with PyTorch for Beginners🐑, 200+ CUDA Kernels, Tensor Cores, HGEMM, FA-2 MMA.🎉

LLM-Viewer. Analyze the inference of Large Language Models (LLMs). Analyze aspects like computation, storage, transmission, and hardware roofline model in a user-friendly interface.

vllm-musa. A high-throughput and memory-efficient inference and serving engine for LLMs

VideoPipe. A cross-platform video structuring (video analysis) framework. If you find it helpful, please give it a star: ) 跨平台的视频结构化（视频分析）框架，觉得有帮助的请给个星星 : )

ThunderKittens. Tile primitives for speedy kernels

developer-roadmap. Interactive roadmaps, guides and other educational content to help developers grow in their careers.

genesis-world. Simulation platform for general-purpose robotics & embodied AI learning.

optimize_softmax. Optimize softmax in triton in many cases

FlashAttention20Triton. Triton implementation of Flash Attention2.0

llava-handbook. 对llava官方代码的一些学习笔记

torchinfo. View model summaries in PyTorch!

ml-engineering. Machine Learning Engineering Open Book

LLaVA. [NeurIPS'23 Oral] Visual Instruction Tuning (LLaVA) built towards GPT-4V level capabilities and beyond.

llama-models. Utilities intended for use with Llama models.

sglang. SGLang is a high-performance serving framework for large language models and multimodal models.

nndeploy. 一款简单易用和高性能的AI部署框架 | An Easy-to-Use and High-Performance AI Deployment Framework

mnn-llm. llm deploy project based mnn. This project has merged into MNN.

awesome-compression. 模型压缩的小白入门教程，PDF下载地址 https://github.com/datawhalechina/awesome-compression/releases

TransformerEngine. A library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit and 4-bit floating point (FP8 and FP4) precision on Hopper, Ada and Blackwell GPUs, to provide better performance with lower memory utilization in both training and inference.

AI-Benchmark. RISCV C and Triton AI-Benchmark

triton-viz. Python

simple_flash_attention. Simple version of Flash-Attention(include V1&V2&Flash-Decoding)

Thinking-Claude. Let your Claude able to think

streaming-llm. Efficient Streaming Language Models with Attention Sinks

Efficient-LLM-Inferencing-on-GPUs. Penn CIS 5650 (GPU Programming and Architecture) Final Project

streaming-llm. [ICLR 2024] Efficient Streaming Language Models with Attention Sinks

flash-attention-minimal. Flash Attention in ~100 lines of CUDA (forward pass only)

FlashAttention. Flash Attention implementation

flashattention2-custom-mask. Triton implementation of FlashAttention2 that adds Custom Masks.

ModelServer. Efficient, Flexible, and Highly Fault-Tolerant Model Service Management Based on SGLang

Liger-Kernel. Efficient Triton Kernels for LLM Training

GPTQ-triton. GPTQ inference Triton kernel

llm-awq. [MLSys 2024 Best Paper Award] AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration

smoothquant. [ICML 2023] SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models

harleyszhang.github.io. 🧗‍♂️ harleyszhang 的个人博客

giscus. A commenting system powered by GitHub Discussions. :octocat: :speech_balloon: :gem:

rope-triton. Python

flashinfer. FlashInfer: Kernel Library for LLM Serving

open-llm-components. A repo showcasing the implementation of various components of open-source LLMs

cuda-course. Cuda

cuda_programming. Code from the "CUDA Crash Course" YouTube series by CoffeeBeforeArch

kernl. Kernl lets you run PyTorch transformer models several times faster on GPU with a single line of code, and is designed to be easily hackable.

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

pytorch-llama. LLaMA 2 implemented from scratch in PyTorch

KuiperLLama. 校招、秋招、春招、实习好项目，带你从零动手实现支持LLama2/3和Qwen2.5的大模型推理框架。

online-softmax. Benchmark code for the "Online normalizer calculation for softmax" paper

FlashAttention20. Get down and dirty with FlashAttention2.0 in pytorch, plug in and play no complex CUDA kernels

eat_algorithm. Data structure algorithm notes and leetcode algorithm solutions and example code.

CUDATutorial. A CUDA tutorial to make people learn CUDA program from 0

cuda-samples. Samples for CUDA Developers which demonstrates features in CUDA Toolkit

triton. Development repository for the Triton language and compiler

LLMRoofline. Compare different hardware platforms via the Roofline Model for LLM inference tasks.

llm_interview_note. 主要记录大语言大模型（LLMs） 算法（应用）工程师相关的知识及面试题

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

resource-stream. GPU programming related news and material links

ollama. Get up and running with Kimi-K2.6, GLM-5.1, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

JARVIS. JARVIS, a system to connect LLMs with ML community. Paper: https://arxiv.org/pdf/2303.17580.pdf

Cpp_Primer_Practice. 搞定C++:punch:。C++ Primer 中文版第5版学习仓库，包括笔记和课后练习答案。

LightLLM. LightLLM is a Python-based LLM (Large Language Model) inference and serving framework, notable for its lightweight design, easy scalability, and high-speed performance.

digital_video_introduction. A hands-on introduction to video technology: image, video, codec (av1, vp9, h265) and more (ffmpeg encoding). Translations: 🇺🇸 🇨🇳 🇯🇵 🇮🇹 🇰🇷 🇷🇺 🇧🇷 🇪🇸

Software-Designer. 软考中级教程-软件设计师

bce-qianfan-sdk. Provide best practices for LMOps, as well as elegant and convenient access to the features of the Qianfan MaaS Platform. (提供大模型工具链最佳实践，以及优雅且便捷地访问千帆大模型平台）

AISystem. AISystem 主要是指AI系统，包括AI芯片、AI编译器、AI推理和训练框架等AI全栈底层技术

awesome-chatgpt-prompts-zh. ChatGPT 中文调教指南。各种场景使用指南。学习怎么让它听你的话。

TensorRT-LLM. TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT LLM also contains components to create Python and C++ runtimes that orchestrate the inference execution in a performant way.

concurrency-programming-via-rust. Rust

rust-course. 什么？你敢放心的把后背交给 AI? 我赌你不敢，那就来学学 AI 时代最酷、最安全、最快的语言吧。本书拥有全面且深入的讲解、生动贴切的示例、德芙般丝滑的内容，这可能是目前最用心的 Rust 中文学习教程 / Book

pydata-book. Materials and IPython notebooks for "Python for Data Analysis" by Wes McKinney, published by O'Reilly Media

Web-Dev-For-Beginners. 24 Lessons, 12 Weeks, Get Started as a Web Developer

nanoPyC. Python

KuiperInfer. 校招、秋招、春招、实习好项目！带你从零实现一个高性能的深度学习推理库，支持大模型 llama2 、Unet、Yolov5、Resnet等模型的推理。Implement a high-performance deep learning inference library step by step

llm_counts. llm theoretical performance analysis tools and support params, flops, memory and latency analysis.

llm-analysis. Latency and Memory Analysis of Transformer Models for Training and Inference

dl_note. 深度学习系统笔记，包含深度学习数学基础知识、神经网络基础部件详解、深度学习炼丹策略、模型压缩算法详解。

Awesome-LLM-Compression. Awesome LLM compression research papers and tools.

fastllm. fastllm是后端无依赖的高性能大模型推理库。同时支持张量并行推理稠密模型和混合模式推理MOE模型，任意10G以上显卡即可推理满血DeepSeek。双路9004/9005服务器+单显卡部署DeepSeek满血满精度原版模型，单并发20tps；INT4量化模型单并发30tps，多并发可达60+。

understand-prompt. 【🔞🔞🔞 内含不适合未成年人阅读的图片】基于我擅长的编程、绘画、写作展开的 AI 探索和总结：StableDiffusion 是一种强大的图像生成模型，能够通过对一张图片进行演化来生成新的图片。ChatGPT 是一个基于 Transformer 的语言生成模型，它能够自动为输入的主题生成合适的文章。而 Github Copilot 是一个智能编程助手，能够加速日常编程活动。

AI-For-Beginners. 12 Weeks, 24 Lessons, AI for All!

RWKV-Runner. A RWKV management and startup tool, full automation, only 8MB. And provides an interface compatible with the OpenAI API. RWKV is a large language model that is fully open source and available for commercial use.

gpt-2-Pytorch. Simple Text-Generator with OpenAI gpt-2 Pytorch Implementation

Prompt-Engineering-Tutior. 🎡 Pompt 提示词工程师入门指南 ~视频字幕+代码资料 [( Python、Golang、NodeJs ) x ( 中文、英文 )]

transformers-bloom-inference. Fast Inference Solutions for BLOOM

Prompt-Engineering-Guide-Chinese. Prompt工程师指南，源自英文版，但增加了AIGC的prompt部分，为了降低同学们的学习门槛，翻译更新

Luotuo-Chinese-LLM. 骆驼(Luotuo): Open Sourced Chinese Language Models. Developed by 陈启源 @ 华中师范大学 & 李鲁鲁 @ 商汤科技 & 冷子昂 @ 商汤科技

Chinese-LLaMA-Alpaca. 中文LLaMA&Alpaca大语言模型+本地CPU/GPU训练部署 (Chinese LLaMA & Alpaca LLMs)

OpenChatKit. Python

transformers. 🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.

DeepSpeed. DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.

Langchain-Chatchat. Langchain-Chatchat（原Langchain-ChatGLM）基于 Langchain 与 ChatGLM, Qwen 与 Llama 等语言模型的 RAG 与 Agent 应用 | Langchain-Chatchat (formerly langchain-ChatGLM), local knowledge based LLM (like ChatGLM, Qwen and Llama) RAG and Agent app with langchain

how-to-optim-algorithm-in-cuda. how to optimize some algorithm in cuda.

TensorRT. NVIDIA® TensorRT™ is an SDK for high-performance deep learning inference on NVIDIA GPUs. This repository contains the open source components of TensorRT.

BELLE. BELLE: Be Everyone's Large Language model Engine（开源中文对话大模型）

annotated_deep_learning_paper_implementations. 🧑‍🏫 60+ Implementations/tutorials of deep learning papers with side-by-side notes 📝; including transformers (original, xl, switch, feedback, vit, ...), optimizers (adam, adabelief, sophia, ...), gans(cyclegan, stylegan2, ...), 🎮 reinforcement learning (ppo, dqn), capsnet, distillation, ... 🧠

MQBench. Model Quantization Benchmark

paper-reading. 深度学习经典、新论文逐段精读

FasterTransformer. Transformer related optimization, including BERT, GPT

llama. Inference code for Llama models

llama.cpp. LLM inference in C/C++

PyTorch-Tutorial-2nd. 《Pytorch实用教程》（第二版）无论是零基础入门，还是CV、NLP、LLM项目应用，或是进阶工程化部署落地，在这里都有。相信在本书的帮助下，读者将能够轻松掌握 PyTorch 的使用，成为一名优秀的深度学习工程师。

SOLOV2_ncnn. the C++ version of solov2 with ncnn

VideosShareByAliyun. 🤪动漫、电视剧、电影、纪录片分享by阿里云盘🫡

awesome-emdl. Embedded and mobile deep learning research resources

hexo-theme-Journal. 隻言片語・於此匯聚 – Moments piled up. Live demo →

md. ✍ WeChat Markdown Editor | 一款高度简洁的微信 Markdown 编辑器：支持 Markdown 语法、自定义主题样式、内容管理、多图床、AI 助手等特性

convnet-burden. Memory consumption and FLOP count estimates for convnets

AI-System. System for AI Education Resource.

MSRA-USTC-AISystemProject-2021. To summit MSRA AI System Project

Cheatsheets. Evolving Cheatsheets for computer languages/software/OS like Python, Numpy, Pandas, Java, Linux, Terminal, Gentoo

Model-Compression-Demo. 模型压缩demo（剪枝、量化、知识蒸馏）

nni. An open source AutoML toolkit for automate machine learning lifecycle, including feature engineering, neural architecture search, model compression and hyper-parameter tuning.

Model-Compression. Model Compression 1. Pruning(BN Pruning) 2. Knowledge Distillation (Hinton) 3. Quantization (MNN) 4. Deployment (MNN)

langchain. The agent engineering platform.

tinynn. A lightweight deep learning library

pytorch-deep-learning. Materials for the Learn PyTorch for Deep Learning: Zero to Mastery course.

ai-edu. AI education materials for Chinese students, teachers and IT professionals.

ultralytics. Ultralytics YOLO26, YOLO11, YOLOv8 — object detection, instance segmentation, semantic segmentation, image classification, pose estimation, object tracking

model-metrics-plot. 🔨🔨🔨(mmplot)used to draw graphs of multiple index parameters such as algorithm accuracy and speed of multiple deep learning models.