MCPcopy Create free account
hub / github.com/ModelEngine-Group/flexai

github.com/ModelEngine-Group/flexai @main

Chat with this repo
repository ↗ · DeepWiki ↗ · + Follow
1,720 symbols 5,119 edges 165 files 303 documented · 18% updated 9d agopreview · 2026-03-03★ 2323 open issues

Browse by type

Functions 1,456 Types & classes 264
What it actually does AI analysis from the code graph — generated when you open this
loading…
README

Flex:ai 介绍及使用说明

概述

Flex:ai是一个面向AI容器场景的开源项目,由上海交通大学、西安交通大学、厦门大学、华为等联合发起。其核心能力包含两大部分,分别是XPU虚拟化和多级智能调度。其中XPU虚拟化分为本地XPU虚拟化和跨节点拉远虚拟化,本地XPU虚拟化支持将1台服务器上的XPU算力卡虚拟化分割出多个虚拟算力单元,实现单张算力卡在多个容器的共享。跨节点拉远虚拟化支持通过RDMA或TCP网络访问远端节点上的算力卡。多级智能调度支持对切分出的虚拟卡进行Binpack等资源级调度,以及对AI训推任务进行分时调度等能力。

动机

用户的AI业务集群通常是大小模型混部的场景,集群除了运行AI大模型如Deepseek、Qwen系列,还有大量的小参数量模型,如CV模型、Embedding模型等,这些小参数量模型无法充分使用整张算力卡的资源,从而导致昂贵的GPU/NPU资源存在浪费。此外算力集群中运行的AI工作负载千差万别,如何基于有限的算力资源对大并发的AI工作负载进行高效调度是一个难题。基于上述的背景,我们构筑了Flex:ai开源项目,提供将XPU算力卡进行虚拟化切分,以及面向AI训推任务和集群资源做智能调度的能力。

Ascend NPU虚拟化

请查看Ascend-Virtual-Service

GPU虚拟化

请查看GPU-Virtual-Service

License

Flex:ai is licensed under the MIT with additional conditions. Please read the LICENSE file for details.

Extension points exported contracts — how you extend this code

Core symbols most depended-on inside this repo

Shape

Method 849
Function 607
Class 161
Struct 64
Enum 15
Interface 12
TypeAlias 12

Languages

C++53%
Go31%
Python17%

Modules by API surface

GPU-Virtual-Service/gpu-remoting/include/hashing/robin_hood.h159 symbols
GPU-Virtual-Service/gpu-remoting/src/client/cudnnHook.cc48 symbols
GPU-Virtual-Service/gpu-remoting/src/client/cudartHook.cc48 symbols
GPU-Virtual-Service/xpu-pool-service/xpu-exporter/common/service/api.pb.go32 symbols
GPU-Virtual-Service/xpu-pool-service/GPU-device-plugin/pkg/api/runtime/service/api.pb.go32 symbols
GPU-Virtual-Service/gpu-remoting/src/client/ncclHook.cc32 symbols
GPU-Virtual-Service/xpu-pool-service/GPU-device-plugin/pkg/gonvml/nvml_wrap.go31 symbols
GPU-Virtual-Service/gpu-remoting/scheduler/requese_handler.py30 symbols
GPU-Virtual-Service/gpu-remoting/src/sample/include/helper_image.h29 symbols
GPU-Virtual-Service/gpu-remoting/include/requestIOV.h29 symbols
GPU-Virtual-Service/xpu-pool-service/GPU-device-plugin/pkg/plugin/xpu/gpu.go26 symbols
GPU-Virtual-Service/xpu-pool-service/GPU-device-plugin/pkg/plugin/util/util.go26 symbols

For agents

$ claude mcp add flexai \
  -- python -m otcore.mcp_server <graph>

⬇ download graph artifact

Ask about this repo answers extend the page