LM Studio - Shared Models

LM Studio Shared LLM Storage Setup (ZimaOS + Windows)

Centralized model storage for GGUF/LLMs across LAN-connected devices using LM Studio.

✨ Overview

This guide shows how to configure LM Studio to use a shared network drive (backed by ZimaOS on a K100 Flash NAS) for storing and resolving downloaded LLM models. This allows multiple Windows machines to use the same models without duplicating downloads.

---

📅 System Configuration Summary

Component	Details
LLM Frontend	LM Studio on Windows 11 Home
NAS OS	ZimaOS on K100 Flash NAS
Storage Setup	3x 256GB SSDs in RAID0 (high-speed shared volume)
Networking	2.5GbE throughout
Windows Drive	Shared drive mounted as Z:

‼️ Future-ready: This setup can be tested with other LLM frontends later, but we're focusing on LM Studio compatibility for now.

---

🔄 Goal

Create a centralized, fast-access model repository on your NAS for use by multiple Windows devices running LM Studio. Models will be downloaded once and stored on a shared ZimaOS volume.

---

⚖️ Step-by-Step Setup Guide

1. 🗂️ Map NAS Shared Drive in Windows

• Open File Explorer
• Map Z: to your ZimaOS NAS volume (e.g. \\ZIMA-NAS\llmshare)
• Enable "Reconnect at login"

2. ❌ Close LM Studio

Ensure LM Studio is not running before proceeding.

3. 📁 Backup Existing Local Model Folder

rename "%USERPROFILE%\.lmstudio\hub\models" models_backup

4. 🔌 Remove the Default Local Models Folder

rmdir "%USERPROFILE%\.lmstudio\hub\models" /s /q

5. ⚙️ Symlink LM Studio Model Path to Shared Drive

mklink /D "%USERPROFILE%\.lmstudio\hub\models" "Z:\"

Expected output:

symbolic link created for ... -> Z:\

6. 🚀 Test It!

• Launch LM Studio

• Download a model (e.g. from TheBloke)

• Confirm it appears in:

  Z:\TheBloke\Mistral-7B-Instruct-GGUF\
  

---

🔧 Tips for Multi-Device Setup

• All machines must:

  • Use the same mapped drive letter (e.g. Z:)
  • Use the same symlink method to redirect LM Studio model path

• Permissions: Ensure shared drive allows read/write access for all users

• Performance: RAID0 over SSDs and 2.5GbE networking ensures minimal bottlenecks

---

🧰 Future Enhancements

• Test shared model repo with other LLM UIs (e.g. LMDeploy, llama.cpp GUIs)
• Add model management scripts (e.g. sync, backup, prune)
• Explore version-locking or hashed storage for CI-like control

---

✅ Confirmed Working

• LM Studio resolves models directly from Z:
• Downloaded models persist and are shared between devices
• No re-download required after machine switch