Pygmalion Guide and FAQ

200,000+ views! Thank you!

Keep Your Smile. Keep on smiling

This guide is WIP, and will be expanded as we get more questions and do more things.

This guide is here to explain how to use the Pygmalion AI language models, as well as to answer questions frequently asked (or soon to be asked) about both the chatbots and the people who make them.

How to use

There are multiple ways to use all our available models:

  1. Use a client such as Agnaistic or connect directly to the Kobold Horde
  2. Set it up locally (requires a bit of technical knowledge).

Using Agnaistic

The easiest way to use our models is to use the Agnaistic client, which can be found here. No registration is required to use it, and your chats will remain anonymous. Agnaistic runs our Pygmalion-6B model on the Kobold Horde, which is a system where people can donate their GPUs for people to run models on for other people to use.

Using the model locally

Several requirements are needed in order to run Pygmalion locally. The larger the Pygmalion model, the more system resources it will take to run it. These requirements are for the models in the precisions they are trained in and with full context length. Requirements are much lower for the quantized versions of these models.

  • Python 3.8 or higher
  • The following Python libraries:
    • torch
    • transformers
    • safetensors
  • The following system RAM requirements:
    • 350M: 2 GB
    • 1.3B: 4 GB
    • 2.7B: 12 GB
    • 6B/7B: 12 GB
    • 13B: 16 GB
  • The following VRAM requirements:
    • 350M: 4 GB
    • 1.3B: 6 GB
    • 2.7B: 10 GB
    • 6B/7B: 16 GB
    • 13B: 32 GB

Further instructions for running Pygmalion locally can be found at our docs page. The docs page in general is a great source of resources, so we recommend checking it out regardless of whether you're running Pygmalion locally or not.


What is this?

Pygmalion is what happened when a bunch of autistic retards from /vt/ and /g/, deprived of freedom by other chatbot services, came together to try to make their own conversational AI. So far, we've released a variety of language models, our current flagship ones being the chat-based Pygmalion-13B model and the instruction-based Metharme-13B model. We hope to release a large variety of even better models in the future.

How can I help out?

We always appreciate more help, whether you have a skillset you think will advance the project further or you just want to support us. You can connect with us on our Matrix, found here:
Alternatively, you can scream at us on /wAIfu/ (/vt/) or /aicg/ (/g/), and if we don't see it, someone else probably will and bring it to our attention.


Pub: 21 Dec 2022 06:15 UTC
Edit: 05 Jul 2023 20:46 UTC
Views: 285722