LM Studio Guide

LM Studio Integration Guide

LM Studio is an excellent tool for running local LLMs with a user-friendly interface. This guide explains how to connect it to RenLocalizer safely.

1. LM Studio Configuration

1. Download & Install: Get LM Studio from lmstudio.ai.
2. Download a Model: We recommend Qwen 2.5 7B Instruct or Llama 3.1 8B Instruct.
3. Start the Server:
   • Click the Local Server icon (the double-headed arrow) on the left sidebar.
   • Select your model and click Start Server.
   • Take note of the port. By default, LM Studio uses 1234.

2. RenLocalizer Setup

Open RenLocalizer Settings and apply these values:

• Translation Engine: Local LLM
• API Key: local (LM Studio doesn't require a real key)
• Base URL: http://localhost:1234/v1 (Make sure to include /v1!)
• Model Name: Enter the exact ID shown in LM Studio (e.g., qwen2.5-7b-instruct)

3. Optimization for Qwen Models

Qwen is highly effective at following syntax instructions. To get the best results:

• Temperature: Set to 0.2 or 0.3.
• Max Batch Size: Start with 5. If your GPU has more than 12GB VRAM, you can try 10.
• System Prompt: Ensure the default localized system prompt is used. It contains critical instructions about preserving [variables] and {tags}.

4. Common Troubleshooting

• Connection Refused: Ensure the server is actually "Started" in LM Studio. Check if a firewall is blocking port 1234.
• Garbage Output: Make sure you are using an "Instruct" or "Chat" version of the model, not a "Base" model.
• Speed Issues: Ensure "GPU Offload" is enabled in LM Studio settings to use your graphics card instead of your CPU.

Why use LM Studio?

• 100% Offline: No data leaves your computer.
• No Limits: No API quotas or monthly fees.
• Censorship Free: Unlike Gemini or OpenAI, you can translate any content as long as the model allows it.