tencentmusic/cube-studio

cube studio开源云原生一站式机器学习/深度学习/大模型AI平台，mlops算法链路全流程，算力租赁平台，notebook在线开发，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，边缘计算，标注平台自动化标注，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，AI模型市场，支持国产cpu/gpu/npu 昇腾生态，支持RDMA，支持pytorch/tf/mxnet/deepspeed/paddle/colossalai/horovod/ray/volcano等分布式

What it solves

Cube Studio is an open-source, cloud-native machine learning platform designed to provide a one-stop shop for the entire ML lifecycle. It addresses the complexity of managing resources, users, and infrastructure for AI development, training, and deployment in a large-scale environment.

How it works

It operates as a cloud-native platform that integrates various infrastructure capabilities. It manages compute resources (CPU, GPU, and specialized AI chips), storage (NFS, S3, etc.), and network configurations. It provides a centralized management interface for project groups, user roles (RBAC), and resource allocation across multiple Kubernetes clusters, including support for edge clusters and serverless modes (Tencent Cloud and Alibaba Cloud).

Who it’s for

It is designed for organizations and teams that need a scalable, enterprise-grade ML platform to manage their AI workloads across diverse hardware and cloud environments.

Highlights

  • Broad Hardware Support: Supports a wide range of GPUs (T4, V100, A100) and domestic AI chips (DCU, NPU, MLU), as well as RDMA and vGPU.
  • Enterprise Management: Includes built-in RBAC, SSO (LDAP, OID), and detailed resource metering and billing for development, training, and inference.
  • Flexible Infrastructure: Supports multiple Kubernetes clusters, containerd, and a variety of distributed storage options (S3, MinIO, CephFS, etc.).
  • Cloud-Native Integration: Offers serverless cluster modes for major cloud providers and supports edge cluster deployments for training and inference.