DKubeX Overview

Attention

This is the user guide documentation for DKubeX v0.8.1. For the user guide rearding DKubeX v0.7.9.1 and v0.7.9.3, please go to the following link provided below:

DKubeX v0.7.9.1

Documentation for DKubeX v0.7.9.1

_images/dkubex_hero.png

DKubeXTM is a Kubernetes-based AI platform supporting two main workflows: LLMOps and MLOps.

LLMOps Features:

  • LLM Catalog: Access to multiple, ready to deploy open-source LLMs.

  • RAG: Build genrative AI applications with your private data.

  • Fine-tuning: Train LLMs with your private data.

  • Serving: Deploy your custom, fine-tuned models on any of the LLMs provided in our catalog.

  • SecureLLM: Monitor requests & responses, collect RLHF data & track security & evaluation metrics.

  • SecureChat: Deploy multiple custom chat applications with any of the LLMs deployed on DKubeX.

MLOps Features:

  • Notebooks: Rapid experimentation with Jupyter, VSCode or R.

  • Pipeline Orchestration: Flyte or Kubeflow for streamlined workflows.

  • Training & Serving: Train & serve traditional ML models.

Common Tools:

  • Custom UI: Manage jobs and clusters through a user-friendly interface.

  • MLFlow: Track experiments and lineage.

  • Logs & Metrics UI: Troubleshoot workflows with custom logs and metrics.

  • Compute Resource Optimization: Optimize training/fine-tuning cost using SkyPilot.

  • Collaboration: Easily share workspaces between users.