Zum Inhalt

Frequently Asked Questions (FAQ)

This FAQ covers the most common questions about DeepExtension — its purpose, usage, licensing, and how it compares to other tools.

What is DeepExtension and who is it for?

DeepExtension is an enterprise-grade platform designed for managing the full lifecycle of large language models (LLMs).
Unlike tools made purely for AI developers, DeepExtension is designed for domain experts, business analysts, and cross-functional teams to collaborate seamlessly with AI developers.

How is DeepExtension different from W&B, Comet, or H2O.ai?

While tools like Weights & Biases, Comet, and H2O focus on experimentation tracking and are built primarily for data scientists, DeepExtension is designed with domain teams in mind — with UI-driven workflows for fine-tuning, evaluation, and deployment, without requiring code or ML expertise.

Is DeepExtension free to use?

Yes. There is a free version that includes all major features and is more than sufficient for individual users or small teams.
Larger teams or enterprises can unlock additional functionality via commercial plans.

Is DeepExtension open-source?

Partially.

  • The AI training components (including adapters, GRPO, MLX training, etc.) are planned to be released under Apache-2.0.
  • The core application (UI + lifecycle backend) is closed-source, but Docker images are publicly available so that everyone can use and deploy DeepExtension without paying.

Do I need to know how to code to use DeepExtension?

No. You can:

  • Train models
  • Evaluate model quality
  • Run inference
  • Use document-based RAG

All from the web UI, with zero coding.

However, for the initial installation (especially on Linux/Mac), someone with basic Docker knowledge should assist.

What AI platforms are supported?

Currently:

  • CUDA (Linux or Windows via WSL) — full support including training
  • MLX (macOS, Apple Silicon) — lightweight training with prebuilt methods

We are actively evaluating more platforms like AMD ROCm, ONNX Runtime, and Windows CPU inference.

Can I fine-tune my own LLMs?

Yes.
DeepExtension supports several parameter-efficient fine-tuning methods like SFT, LoRA, and GRPO, with no coding required.
You can also implement your own training methods if you have AI experience.

Can I run DeepExtension offline?

Yes.
The entire platform can run locally or inside a private cloud, without any external API dependencies.
You can use local base models, run training, deploy models, and even evaluate them offline.

How do I install DeepExtension?

We provide detailed installation instructions for: - CUDA（Linux or Windows via WSL) systems - macOS (Apple Silicon)

Installation uses Docker Compose, and a typical setup can be completed in under 30 minutes if requirements are met.
See Installation Guide for details.

Still have questions? Contact us via the Support Page.