LLM Mapper Step

The LLM Mapper step is to get Large Language Model results based on the prompt provided.

Praise Magidi avatar
Written by Praise Magidi
Updated over a week ago

It is used to generate Large Language Model (LLM) results based on prompts provided. It works with an integrated AI GPT model to give some information, similar to how Chat GPT works.

The distinction is that the model within Synatic is not publicly hosted/exposed in any way. It does not consume nor hold any client data. This makes it secure for use and ensures client data is not shared publicly.

For more information on Large Language Models, click on the link below:

Illustrated in the below image is a flow where the Record Generator is used as a Source Step. The LLM Mapper step provides a prompt. The Buffer step is used to store the data.

Adding a LLM Mapper Step

To add a LLM Mapper step in a flow, follow the below instructions:

1. Click on the highlighted icon as shown in the below image.

2. The below page will appear. Select or search for a step to pull out the source data. Click on the Mapper tab and select LLM Mapper as shown in the image below.

Configuring a LLM Mapper Step

Illustrated in the below image is the configuration that is available in the LLM Mapper step.

Model – The model being used. GPT 3.5 and GPT4 are currently available options.

Prompt – The prompt to send to the LLM for processing. This field supports handlebars syntax. For example, the prompt could refer to a record in the Record Generator.

From Path – Field in the chat result to map the record. Leave blank to map the entire response.

Temperature – Controls the accuracy and creativity of the response. It can only be a value between 0 and 1. A lower value gives a higher amount of accuracy. This is set at 0.2 by default.

To Field – Field to map on the unified record. Will be added to the property record.

Using a LLM Mapper Step

There are 2 ways to prompt the field:

1. The prompt field in the LLM Mapper step:

In the example below, the prompt has been provided.

The image below shows the output of the LLM Mapper step.

The message section shows the content from the prompt being successfully written. The index and prompt filter results show metadata from the response with different ratings such as hate, self-harm etc. The usage section shows the number of tokens that were sent in and out.

2. Referring to a record in the prompt field of the LLM Mapper step:

The image below shows the Record Generator step. This address will be formatted using the LLM Mapper step prompt field.

The image below shows the LLM Mapper step. The prompt has been provided in the field to format the address provided in the Record Generator.

The image below shows the output of the LLM Mapper step. The address has been formatted as required.

Did this answer your question?