A lightweight alternative to OpenClaw with enhanced mobile control

AI agents running research on single-GPU nanochat training automatically

NVSHMEM‑Tutorial: Build a DeepEP‑like GPU Buffer

Preview Code for Continuum Paper

A compact implementation of SGLang, designed to demystify the complexities of modern LLM serving systems.

Based on Nano-vLLM, a simple replication of vLLM with self-contained paged attention and flash attention implementation

A framework for efficient model inference with omni-modality models

The Intelligent GUI Agent for Mobile Phones

The wafer-native AI accelerator simulation platform and inference engine.

Implementation of "Interleaved Latent Visual Reasoning with Selective Perceptual Modeling".

Ray tutorials from Anyscale

集成式幻觉检测系统

🚀🚀 「大模型」2小时完全从0训练64M的小参数GPT！🌏 Train a 64M-parameter GPT from scratch in just 2h!

The best ChatGPT that $100 can buy.

verl-agent is an extension of veRL, designed for training LLM/VLM agents via RL. verl-agent is also the official code for paper "Group-in-Group Policy Optimization for LLM Agent Training"

Understanding R1-Zero-Like Training: A Critical Perspective

A Gym for Agentic LLMs

The Intelligent GUI Agent for Mobile Phones

to learn RL

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

to learn Minimalistic 4D-parallelism distributed training framework for education purpose

Easiest and laziest way for building multi-agent LLMs applications.