Chat AI: Run LLMs with your own hardware

Chat AI is an interface that sits on top of a custom engine which is responsible for processing requests and responses to and from an active Google T5-Flan model.

Bundled Installation

Official Build can be acquired here for those who want to use a compiled version of Chat AI without having to install any additional requirements

Development

Prerequisites

• Ubuntu 20.04+ or Windows 10+
• Python 3.10.6
• pip-23.0.1

Pypi installation

Use this installation method if you intend to use Chat AI from the command line or with other python libraries or scripts.

Windows

pip install torch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 --index-url https://download.pytorch.org/whl/cu117
pip install aihandlerwindows
pip install https://github.com/w4ffl35/diffusers/archive/refs/tags/v0.14.0.ckpt_fix.tar.gz
pip install https://github.com/w4ffl35/transformers/archive/refs/tags/tensor_fix-v1.0.2.tar.gz
pip install https://github.com/acpopescu/bitsandbytes/releases/download/v0.37.2-win.0/bitsandbytes-0.37.2-py3-none-any.whl
pip install chatai --no-deps

Linux

pip install https://github.com/w4ffl35/diffusers/archive/refs/tags/v0.14.0.ckpt_fix.tar.gz
pip install https://github.com/w4ffl35/transformers/archive/refs/tags/tensor_fix-v1.0.2.tar.gz
pip install chatairunner

Development installation

Use this installation method if you intend to modify the source code of Chat AI.

• Ubuntu 20.04+ or Windows 10+
• Python 3.10.6
• pip-23.0.1

1. Fork this repo on github
2. git clone https://github.com/Capsize-Games/chatai
3. python -m venv env - skip if you don't want to use venv
4. cd chatai && pip install -e .
5. cd chatai && python main.py

Here are the special tokens used by the model:

• : Used to pad sequences to a fixed length.
• : Beginning of sequence token, used to indicate the start of a sequence.
• : End of sequence token, used to indicate the end of a sequence.
• : Unknown token, used to represent out-of-vocabulary words.
• <extra_id_X>: Special token used to represent additional task-specific labels, where X is a number from 0 to 99.
• : End of turn token, used in conversation models to indicate the end of a speaker's turn.