Nano vLLM

Nano vLLM

Fast and memory-efficient exact attention

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+。

Transformer: PyTorch Implementation of "Attention Is All You Need"

Transformer: PyTorch Implementation of "Attention Is All You Need"

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

hpc 教程，包含集合通信(mpi、nccl)、cuda 编程、向量化 SIMD、RDMA 通信等

hpc 教程，包含集合通信(mpi、nccl)、cuda 编程、向量化 SIMD、RDMA 通信等

Best practices & guides on how to write distributed pytorch training code

Efficient Distributed GPU Programming for Exascale, an SC/ISC Tutorial

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

Examples from Programming in Parallel with CUDA

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

Qwen3 is the large language model series developed by Qwen team, Alibaba Cloud.

TTRL: Test-Time Reinforcement Learning

Large-scale LLM inference engine

Quantized Attention that achieves speedups of 2.1-3.1x and 2.7-5.1x compared to FlashAttention2 and xformers, respectively, without lossing end-to-end metrics across various models.

Quantized Attention achieves speedup of 2-5x and 3-11x compared to FlashAttention and xformers, without lossing end-to-end metrics across language, image, and video models.

Yan (炎) is a high-performance CUDA operator library designed for learning purposes while emphasizing clean code and maximum performance.

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

Kimi-VL: Mixture-of-Experts Vision-Language Model for Multimodal Reasoning, Long-Context Understanding, and Strong Agent Capabilities

Implementing DeepSeek R1's GRPO algorithm from scratch

Video-R1: Reinforcing Video Reasoning in MLLMs [🔥the first paper to explore R1 for video]

Rethinking RL Scaling for Vision Language Models: A Transparent, From-Scratch Framework and Comprehensive Evaluation Scheme

Rethinking RL Scaling for Vision Language Models: A Transparent, From-Scratch Framework and Comprehensive Evaluation Scheme

Video-R1: Reinforcing Video Reasoning in MLLMs [🔥the first paper to explore R1 for video]