How To Build a Custom Chat LLM on Your Data

This is one of the fastest ways to build a custom ChatGPT-like system on top of your data.

It's called ChatLLM (by Abacus.AI )

Here is a demo of how to build a simple custom chat LLM:

Step 1

From within the Abacus.AI platform, I chose to create a new project using the ChatLLM options.

This will allow you to build a custom LLM chat system.

As you can see, there are also many other use cases supported.

Step 2

Then you attach a dataset or as many as you want.

Data is not perfect, so you can perform transformations through feature groups that will be used by the chat LLM to answer questions.

We are using a TV series dataset and combined some columns into a text field:

Step 3

The next step is to train a model.

You configure the documents you want to use, the evaluation (if any), the chat model, and the text processing steps.

Some orchestration will happen here but we don’t need to worry about that. It's all taken care of for us.

Step 4

Once the model is fully trained, we can finally deploy it!

The whole thing took me less than an hour to train and deploy.

I didn’t need to code anything! I just used the wizards to configure the pipeline.

Step 5

With our custom chat LLM deployed, you can now interact with it as you would with ChatGPT.

We are using a tv series dataset that we downloaded from Kaggle.

We can ask it any question and get responses along with scored docs.

And here is a preview where I test the chat LLM:

The model will obviously not be perfect the first time you train it.

And there might be new data you want to retrain on. That's supported by Abacus.AI .

I used the built-in AI chat to transform data and retrain the model. I'm really impressed by ChatLLM.