CUDA Templates and Python DSLs for High-Performance Linear Algebra

Cosmos-RL is a flexible and scalable Reinforcement Learning framework specialized for Physical AI applications.

VeOmni: Scaling Any Modality Model Training with Model-Centric Distributed Recipe Zoo

Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

基于LLaMA2-7B增量预训练的藏文大语言模型TiLamb（Tibetan Large Language Model Base）

PyTorch native quantization and sparsity for training and inference

A Throughput-Optimized Pipeline Parallel Inference System for Large Language Models

Train speculative decoding models effortlessly and port them smoothly to SGLang serving.

讨贼王云鹤檄文

🚀🚀 「大模型」2小时完全从0训练26M的小参数GPT！🌏 Train a 26M-parameter GPT from scratch in just 2h!

My learning notes for ML SYS.

FlashMLA: Efficient Multi-head Latent Attention Kernels

Implement Flash Attention using Cute.

A PyTorch native platform for training generative AI models

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

中文的C++ Template的教学指南。与知名书籍C++ Templates不同，该系列教程将C++ Templates作为一门图灵完备的语言来讲授，以求帮助读者对Meta-Programming融会贯通。(正在施工中)

Meta Lingua: a lean, efficient, and easy-to-hack codebase to research LLMs.

《Machine Learning Systems: Design and Implementation》- Chinese Version

Course to get into Large Language Models (LLMs) with roadmaps and Colab notebooks.

Excalidraw-CN 是支持中文手写和多画布的 Excalidraw 白板工具。Excalidraw-CN is a whiteboard supporting Chinese hand draw font and multi-canvas based on Excalidraw.

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

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

🤗 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.

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

hpc-learning