Skip to main content
Version: 0.1.0

Key Features

This page introduces the key features of the NuFi platform.

NPUOps — From Existing MLOps to NPU Serving​

NuFi does not replace your training infrastructure. It brings in models trained in existing MLOps environments such as Kubeflow and MLflow, and handles NPU porting (compilation → model registry storage → validation) through serving deployment. MLflow integration lets you pull models directly from the model registry, so you can keep your existing MLOps investment while extending into NPU serving.

Supported devices: GPU (NVIDIA), RNGD (FuriosaAI). Devices not on the list can still be managed through manual registration.

Lab — If You Don't Have an Existing MLOps​

Even without an existing MLOps environment, you can use NuFi Labs as your training environment. Create web-based development environments such as Jupyter Notebook, VS Code, and LlamaFactory with a single click, and serve models trained in a Lab right away.

See Set up a development environment for details.

Public-Sector-Grade Security — RBAC · Audit Log​

NuFi alone can meet the strict security requirements of public sector, finance, and defense domains.

  • Role-Based Access Control (RBAC): Integrates with the Keycloak authentication system to strictly separate permissions by role. Sensitive admin features are restricted to designated infrastructure administrators, blocking misuse by general users at the source.
  • Audit Log: Every action — who, when, what — is recorded. During security audits you can submit access records immediately, satisfying the information security compliance requirements of public institutions.

NPU vs GPU Performance/Watt Comparison Dashboard​

Validate NPU adoption benefits with quantitative data. Serve the same model on both an NPU and a GPU, and compare core metrics including power efficiency (Performance/Watt) directly in the dashboard.

Integrated Hardware Monitoring​

From the cluster down to individual devices, observe everything inside the platform UI.

  • Cluster-wide and per-node CPU · memory · storage status
  • Per-device utilization, temperature, power, and memory visualization
  • Devices in the "In Use" state directly show the occupying Pod name — instantly trace the source when an incident occurs

See Monitoring for details.