Sukii / jetmoe

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Python scripts for running JetMoE on CPU/GPU

Please pull jetmoe-8b model from Huggingface to model/ folder and jetmoe-8b-chat model from Huggingface to chat-model/ folder by running:

./install.sh

It should take about 15 minutes to pull both the models.

Now you should be able to run:

python jet.py

and

python chat-jet.py

Please use higher version of python (>=3.10) and higher version of transformers (>=3.11).

You need atleast 64GB RAM and about 10 CPU cores to be able to run it. But if you have GPUs then it should work much more efficienty as it can process all those large arrays together through those thosands of parallel-gears of GPUs.

The bash script for pulling models should work directly in Linux/MacOSX and in Windows if you use install git-bash terminal (https://gitforwindows.org/).

About


Languages

Language:Python 54.1%Language:Shell 45.9%