# data-infra/cube-studio

cubestudio开源云原生一站式机器学习/深度学习/大模型AI平台/MaaS/mlops/人工智能平台/训推平台，算法全链路流程，多租户，算力租赁平台，token中转，拖拉拽任务流pipeline编排，多机多卡分布式训练，超参搜索，推理服务，VGPU虚拟化，云边端协同，边缘计算，自动化标注平台，deepseek等大模型sft微调/奖励模型/强化学习训练，vllm/ollama/mindie大模型多机推理，私有知识库llmops智能体，AI模型市场，支持国产异构算力调度,昇腾/寒武纪/海光/摩尔/沐曦等，支持ib/roce/RDMA，信创支持

Repository: https://github.com/data-infra/cube-studio
Canonical: https://ross.abutalabs.com/products/data-infra-cube-studio
Homepage: https://cubestudio.vip
Language: Python
License: NOASSERTION
License Family: other
Topics: ai-platform, automl, cube-studio, deepseek, kubernetes, llmops, mlops, pipeline, vgpu, workflow, ascend, inference, npu, maas, machine-learning, cubestudio
Last push: 2026-08-17T01:22:20+00:00

## Health v2 (maintenance only)
Score: 80/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 98, release rhythm 71, longevity 56
- inputs: {"age_days": 792, "days_push": 17, "days_rel": 38, "gap_med": 92.0, "n_releases_24m": 9}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2448, forks 196 (observed 2026-08-28T04:06:52.709662+00:00)

## What it is
CubeStudio is an open-source, cloud-native, all-in-one AI platform covering the full machine learning lifecycle (MLOps/MaaS/LLMOps), including multi-tenant compute scheduling, drag-and-drop pipeline orchestration, distributed training, hyperparameter search, and LLM fine-tuning/inference. It runs on Kubernetes, supports vGPU virtualization, RDMA networking, edge clusters, and domestic heterogeneous accelerators (Ascend, Cambricon, Hygon, Moore Threads, MetaX) with private/offline deployment.

## Use cases
- deploy a self-hosted mlops platform on kubernetes
- orchestrate training pipelines with drag-and-drop
- run multi-node multi-gpu distributed training
- fine-tune llms like deepseek with sft or rlhf
- serve llm inference with vllm or ollama
- schedule and rent gpu/npu compute across teams
- build a private knowledge base rag chatbot
- automate multimodal data annotation

## When to choose
- you need a full-chain AI platform covering data, training, and inference in one self-hosted deployment
- your organization uses domestic heterogeneous accelerators (Ascend, Cambricon, Hygon) or requires xinchuang/offline deployment
- you need multi-tenant compute scheduling, quota billing, and compute rental
- you want drag-and-drop pipeline orchestration plus distributed training and hyperparameter search
- you need LLM fine-tuning, multi-node inference, and private RAG agents in one platform

## When to avoid
- you only need a lightweight experiment tracking or model registry tool rather than a full platform
- you cannot operate a Kubernetes cluster or lack ops capacity for platform maintenance
- you need a managed cloud service with vendor support instead of self-hosted software
- your workload is a single small model with no need for multi-tenancy or scheduling

## Facets
- artifact type: application
- maturity: active
- function: machine-learning, deep-learning, llm-inference, llm-training, rag, agent-framework, workflow-automation, scheduling, monitoring, data-visualization, etl, container-orchestration, web-framework, self-hosted
- domain: machine-learning, deep-learning, large-language-models, artificial-intelligence, gpu-computing, self-hosted, cloud-computing, microservices
- platform: self-hosted, python, cross-platform
- tags: mlops, maas, llmops, ai-platform, vgpu, rdma, pipeline-orchestration, distributed-training, hyperparameter-search, inference-serving, multi-tenant, domestic-accelerators, ascend-npu, edge-computing, compute-scheduling, model-marketplace, annotation-platform, fine-tuning, reinforcement-learning, xinchuang, data-engineering, containers, kubernetes, docker, web-server, linux, gpu

## Member repositories
- data-infra/cube-studio (main) score 80

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:52.709662+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:30:27.978147+00:00, confidence not recorded.
  - readme: https://github.com/data-infra/cube-studio (fetched 2026-08-28T04:06:52.709662+00:00, sha f004defec130)
  - homepage: https://cubestudio.vip (fetched 2026-08-29T10:12:00.653635+00:00, sha eeee2e074962)
  - site_page: https://www.cubestudio.vip/docs/00-%E7%B4%A2%E5%BC%95/%E5%85%A8%E6%96%87%E7%9B%AE%E5%BD%95 (fetched 2026-08-29T10:12:00.656239+00:00, sha 3190c382b98e)
- Data as of 2026-08-30T08:39:29.467469+00:00.