Alright, so the current guide available is very basic at best... so I'm going to walk you n00bs through this ♥♥♥♥. No, I will not post links to software, you're going to have to get used to looking up some basic information and we might as well get you ready for it now.

First thing's first. Go download LM Studio (That's right... look it up.) LM Studio contains a full search function that will allow you to download LLM models directly from HuggingFace, and you can easily load in and test models before you import them to ensure they work correctly. It will also contain model information and parameters that you will need when importing the LLM in ChatWaifu.

*Once LM Studio is installed, open it.

*Under the "My Models" tab you will see a field to set the path for your "Models Directory". Set this to whatever you would like.

When searching for models, you can click the name to display an "information card". Take note of the exact number listed in the "parameters" field. (For example, an 8B model may actually be 8.1B parameters.)

After choosing and testing your model/s, use the LLM tab in Waifu to import them from your LM Studio "model" folder. After you choose to import, you can use the following template to input the required information. In the example, we'll be using marco-o1-uncensor by 'mradermacher'.

Display Name - You Decide

LLM Name - marco-o1-uncensor (This needs to appear exactly as it does in the card. The easier way to get it is from the file path when importing. For example: C:\Users\realc\.ollama\models\blobs\mradermacher\L3-15B-Stheno-v3.3-32K-exp-GGUF\L3-15B-Stheno-v3.3-32K-exp.Q4_K_S.gguf You would use the name BEFORE the part of the file path containing "L3-15B-Stheno-v3.3-32K-exp.Q4_K_S.gguf", which would be "L3-15B-Stheno-v3.3-32K-exp-GGUF". You would use this example for any LLM you import.)

GGUF File - link to LLM on your drive. (Waifu will make a copy of this file and create a fileset it can load with ollama in Waifu. Also keep in mind this will make a 2nd copy of the LLM on your disk, be aware of your storage space.)

Param Count - 7.6 Billion (Can be found using LM Studio in the model's information.)

Requirement - 6 (This is GB of RAM needed to load in the model. If unsure, give it too much then check the info. after it loads in with
Waifu, reload and set correctly.)

Leave the rest alone if you don't know what you're doing.


SUGGESTED LLMs & USAGE INFORMATION:
Here I'll just go over a few basic things before listing a few I personally like. First, you want at least an 8GB GPU with 32 GB of ram. You can run them as basic "chat bots" on this app with a 3B but they suck. You want to run at least an 8B L3-3.3 Instruct model. Q4KS/Q4KM/Q08, order is based on quality with Q4KS being the lower end. You can try out the other quants, but anything below the KS can be mediocre.

On an 8GB GPU with at least 32GB of ram, you should be able to load in most models up to around 25-30Bs. GPU size will affect processing and generation speed, RAM will affect your ability to load the model in and have it ready for the GPU. If you're short of GPU, you're CPU will be able to cover up to a certain extent. Again, testing is key. Don't even bother with the 70B+ models unless you're sitting on a 40GB+ GPU with more RAM than god. Models listed with higher "window context limits"(usually in increments of 4K, 8K, 16K, and so on), generally have a longer "memory".

Look for "Conversational" LLMs, generally those that specialize in roleplay. Instruct models will be more adept at using "call tools", but you'll probably have to explain it to them. You can look at any of my 'cards' on the workshop to get an idea of how to write character prompts. I'll post one more guide explaining how to format a character prompt correctly, for now... the LLM list. Most LLMs will require you to explain things to them and set rules. That guide will tell you how.

marco-o1-uncensor (8B, good at call tools. Not bad at roleplay. Up to 32K context memory.)

L3-Stheno-v3.2-12.2B-Instruct-GGUF (A slightly smaller version of the LLM below. Lower context, 8K I think. Maybe try it if the 15B doesn't work.)

L3-15B-Stheno-v3.3-32K-exp-GGUF (15B unit. Will eat up about 16-17 gigs of RAM. One of the best RP models you can download right now without requiring beefy computer. Up to 32K context memory.)

And that's it, no more holding your hand. (I know where it's going to have been after this)

Make sure you check out the other guide about writing templates if you want to make your own characters.

You can also use Blender v4.2LTS (available on Steam, set version in properties, under "betas") with MMD Tools (mikomikodance, an old .pmx file editor) to create and edit your own models. I'm not writing a walkthrough for that ♥♥♥♥ though... still learning myself and way too much information.