Cube Studio
An archived, open-source, cloud-native machine learning and large-model AI platform covering the full MLOps pipeline, from notebook development to distributed training and inference serving.
Cube Studio, per its GitHub description, offered a compute-rental platform, online notebook development, drag-and-drop pipeline orchestration, multi-machine multi-GPU distributed training, hyperparameter search, VGPU-virtualized inference serving, edge computing, an automated labeling platform, SFT, reward-model and reinforcement-learning fine-tuning for large models such as DeepSeek, multi-machine inference through vLLM, Ollama or MindIE, a private knowledge base, and an AI model marketplace.
It was built to run distributed training across PyTorch, TensorFlow, MXNet, DeepSpeed, Paddle, ColossalAI, Horovod, Ray and Volcano, to support domestic CPU, GPU and NPU hardware including the Ascend ecosystem and RDMA networking, and to orchestrate all of this on Kubernetes with Argo and Kubeflow.
The repository was archived on 2026-05-18 and is no longer maintained, with development redirected to a successor repository named in the material; its license is not asserted.