jl-codes

jl-codes / dgx-spark-ai

Public

Run GPT-OSS 120B locally on NVIDIA DGX Spark. Fine-tune your own models. Use Cline CLI with zero cloud dependency.

12
3
100% credibility
Found Mar 27, 2026 at 12 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Shell
AI Summary

This repository offers a detailed tutorial and ready-to-use scripts for running massive AI models locally on NVIDIA DGX Spark hardware, turning it into a private AI coding assistant accessible via familiar chat interfaces.

How It Works

1
📰 Discover the Desk AI Guide

You find a friendly step-by-step guide to bring a huge thinking AI right to your powerful desk computer.

2
🔍 Check Your Super Setup

You quickly peek at your computer's brain and memory to make sure everything is powerful enough.

3
⚙️ Prepare Your AI Space

You follow easy steps to ready your computer workspace so the AI can live there comfortably.

4
🚀 Wake Up Your AI Friend

With one simple command, your personal giant AI brain comes alive on your desk.

5
💬 Have Your First Chat

You send a hello message and get a smart reply back, feeling the magic of private thinking power.

6
🔗 Link Your Coding Pal

You connect a handy coding helper tool to chat with your AI anytime while building projects.

🎉 Enjoy Endless Private AI Help

Now your desk supercomputer powers unlimited AI chats and code magic, all yours forever with no outside help needed.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 12 to 12 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is dgx-spark-ai?

This Shell project turns your NVIDIA DGX Spark AI computer into a local AI supercomputer, serving 120B-parameter GPT-OSS models via vLLM's OpenAI-compatible API on localhost:8000. You get quick-start commands like `make serve` to load models, `make train` for QLoRA fine-tuning, and Cline CLI setup for AI coding assistance—all zero cloud dependency. It handles inference, training, and production services on Ubuntu with CUDA 13.0.

Why is it gaining traction?

DGX Spark AI's unified 128GB memory lets you run massive models like GPT-OSS 120B that need multiple GPUs elsewhere, with eager mode fitting them snugly. Makefile targets simplify serving, switching models, health checks, and systemd auto-restart—far easier than manual vLLM tweaks. Cline integration delivers local GitHub Copilot-style coding without API costs or limits.

Who should use this?

ML engineers testing DGX Spark AI price and supercomputer availability for local training on custom datasets. Developers with a DGX Spark AI mini PC or workbench wanting to run GitHub Copilot locally via Cline, or fine-tune coding models offline. Ideal for privacy-focused teams ditching cloud inference.

Verdict

Grab it if you own DGX Spark hardware—polished docs and ready-to-run scripts make the 12 stars and 1.0% credibility score forgivable for an early project. Skip unless you're evaluating local 120B runs; maturity shows in production features, not tests.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.