# The Right Way to Train Models

There are so many confusions on how you can train your AI models and what's the right way. So, here we resolve all your doubts at once : 

<figure><img src="https://26047275-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fz7k4QvTbfR7aTAdgPmkt%2Fuploads%2FoLIjQ3BlrgGNQLkuR90Y%2FScreenshot%202023-06-11%20at%201.35.27%20PM.png?alt=media&#x26;token=5cf4f2c1-d894-4690-bc82-6cf8d3ac7018" alt=""><figcaption></figcaption></figure>

### The Basics

Training AKA *creating a new model* involves 5 input fields to be filled out by you. Firstly, you need to name the model. The default name format is **DEMO NAME (.) SUPERVISED (.) DATA** that you can change to anything.&#x20;

However, we suggest to just change the **DEMO NAME** part as it gives better filtration options later while building apps on these models.

### The License&#x20;

This is where you can decide how you want your models to be utilized by other developers and companies.&#x20;

We have a dedicated section where we have talked about licenses on Supervised. So, find more informations there.&#x20;

### Selecting A Base Model

Here, you have to select the foundational model of your AI. Choose between Ada, Babbage, Curie, and DaVinci.&#x20;

You can find more info about which model to choose from [here](https://platform.openai.com/docs/guides/fine-tuning).&#x20;

### Final Goal for Model

The final goal for model is a definition of what your model is being build for. Each selection can change the way your models responds to conversations.&#x20;

### Dataset

Datasets are the heart of your Supervised LLMs (models). Your data needs to be correctly formatted in order for the model to work in the right way. &#x20;

A sample dataset format is provided can be downloaded from the training page. Make sure your data is in the same format.&#x20;

**For precision, your dataset needs atleast 3500 rows of unique data.**&#x20;