Train the Stuff and Stuff
Clone this repo somewhere, the easiest way is to click the green button that says code on the main repo:
Copy that command to a console and clone the repo somewhere.
Install python on your machine. I'd recommend using https://chat.openai.com/ if you aren't sure how to do this. Make sure to install python with pip.
run pip install -r requirements.txt in the directory you cloned earlier. You will see some stuff about things installing.
Go to huggingface and grab a model, for the purpose of this demo lets use airoboros-1.4)
Click the 3 dots on this page and select clone repo
You'll see this:
If you don't have git lfs installed, follow the instructions on the popup to install git lfs. Once lfs is installed, go to the models directory and:
run git lfs install
then git clone https://huggingface.co/jondurbin/airoboros-7b-gpt4-1.4
This will download the model from huggingface. Once it's done downloading that's it. You did the thing.
https://github.com/CoffeeVampir3/ez-trainer/wiki/Datasets-Overview
Place datasets in the dataset folder. Waow!
To run the thing, navigate to the directory you cloned. It is the thing with train_module.py Run the thing
python train_module.py
It will start a gradio session, navigate to the url in the console output to get to the configuration server, it will very likely be:
Running on local URL: http://127.0.0.1:7860
This is a local gradio server that will run the training session.
A brief overview of the settings here, note that this is not a guide on how to select parameters, only a general overview of what they are: https://github.com/CoffeeVampir3/ez-trainer/wiki/Settings
See one of the examples for some idea of a starting point for training, this is not a process that can be explained simply or briefly, so for serious usage I'd deeply suggest seeking outside wisdom on machine-learning best practices.
That's it, load the stuff and hit the button. This is not a quick process, your console will log the progress of training or any errors that might occur. Good luck!