My learning notes/codes for ML SYS.

Nano vLLM

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

Efficient Triton Kernels for LLM Training

Code repository for the paper - "Matryoshka Representation Learning"

An open-source solution for full parameter fine-tuning of DeepSeek-V3/R1 671B, including complete code and scripts from training to inference, as well as some practical experiences and conclusions.…

An open-source AI agent that brings the power of Gemini directly into your terminal.

🧑‍🚀 全世界最好的LLM资料总结（语音视频生成、Agent、辅助编程、数据处理、模型训练、模型推理、o1 模型、MCP、小语言模型、视觉语言模型） | Summary of the world's best LLM resources.

Text-audio foundation model from Boson AI

Tiny-FSDP, a minimalistic re-implementation of the PyTorch FSDP

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

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

PyTorch bindings for CUTLASS grouped GEMM.

Collect the awesome works evolved around reasoning models like O1/R1 in visual domain

量化代码

📚 从零开始的大语言模型原理与实践教程

Official repository of 'Visual-RFT: Visual Reinforcement Fine-Tuning' & 'Visual-ARFT: Visual Agentic Reinforcement Fine-Tuning'’

DeerFlow is a community-driven Deep Research framework, combining language models with tools like web search, crawling, and Python execution, while contributing back to the open-source community.

pix2tex: Using a ViT to convert images of equations into LaTeX code.

Matrix-Game 2.0: An Open-Source, Real-Time, and Streaming Interactive World Model

An Open-source RL System from ByteDance Seed and Tsinghua AIR

verl: Volcano Engine Reinforcement Learning for LLMs

Awesome RL-based LLM Reasoning

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

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

Production-tested AI infrastructure tools for efficient AGI development and community-driven innovation

DeepEP: an efficient expert-parallel communication library

R1-onevision, a visual language model capable of deep CoT reasoning.