The Toolbar feature in ChatLLM allows users to enhance their custom ChatLLM with additional functionalities such as Image Generation, Doc Generation, and more.
This field is optional and only required if you want your custom ChatLLM to showcase a toolbar.
You can control the context of the LLM to use for the toolbar as follows:
Enable Web Search:
Allows the LLM to perform web search requests and fetch pages when responding to questions that require access to recent or specific information. Not all questions will result in web requests, only those that the model determines will benefit from access to information. This initial determination step can add some additional latency to responses. Enabling this option may send data to alternate models. Do not enable this option if it is necessary to strictly control the models used.
Include General Knowledge:
If you have attached a Doc Retriever, it would allow the LLM to access information from its training data to answer queries. Currently supports only Unstructured Data, doesn't support Structured Data.
The Toolbar offers a variety of options to enhance your ChatLLM experience:
| Option | Description |
|---|---|
| Video Analysis | Analyzes video content to extract insights or data. |
| Scrape URL | Extracts information from a specified web URL. |
| Video-Gen | Generates video content based on input parameters. |
| Humanize | Converts GenAI text into something that is more human-like. |
| Doc-Gen | Creates documents from provided data or templates. |
| Image | Handles image-related tasks such as editing or generating images. |
| Code | Assists with coding tasks, including writing and debugging code. |
| Playground | An interactive space for testing and experimenting with code or ideas. |
| Editor | A tool for editing text or documents. |
| Screenshot | Captures and manages screenshots. |
| Powerpoint-Gen | Generates PowerPoint presentations from data or templates. |
Behavior Instructions are directives that shape how the model approaches the problem to be solved. They are used to direct the choices and methods used by the LLM during the process of getting the answer. These instructions become more specific to the use case you are tackling.
Some examples for the different scenarios can be:
Answering Questions on a Document:
To make the LLM perform a more focused query, you can say:
"Use only the top search results while generating the answer."
Querying a Database:
To make the LLM provide more background on the retrieved rows, you can say:
"While writing SQL code try to retrieve related columns to provide more background to the results."
Using Web Search:
To force the LLM to always perform a web search, you can say:
"Always perform a web search before answering the query."
Response Instructions focus on the way the LLM responds to the user setting. They can be used to control format, persona, tone, as well as the structure of the answers that the LLM generates.
Here are a subset of aspects that you can control with response instructions:
Tone:
Directing the model to be formal, casual, humorous, or empathetic.
Persona:
Instructing the model to behave like a specific character or role, such as a technical expert, a friendly assistant, or a strict moderator.
Ethical Guidelines:
Ensuring the model avoids certain topics, respects privacy, or adheres to community guidelines.
Engagement:
Guiding the model to take initiative in the conversation, ask follow-up questions, or provide detailed explanations.
Length:
Specifying the desired verbosity of the model's responses, from concise to detailed.
Complexity:
Adjusting the complexity of the language used, which can be important for different audience levels.
Content:
Directing the model to include or exclude certain types of information, such as examples, analogies, or references.
Clarification:
Instructing the model on how to handle ambiguous or incomplete queries, such as asking for clarification or making educated guesses.
An example instruction would be:
"You are a professional consultant. Keep responses under three sentences unless elaborating on complex topics.
Use bullet points for step-by-step instructions. Always confirm user understanding before moving on to the next step."
You can enable the Toolbar feature during the training or re-training step of chatbot creation.
Similarly, please select "No" under Enable Tool Bar to disable this feature.
Before enabling Web Search, consider the following:
Enabling this option may send data to alternate models.
Ensure that it aligns with your data privacy and security requirements.
Be aware that enabling Web Search can introduce additional latency in responses.