Generative AI applications have taken over the world. Many tools, such as ChatGPT or Claude, help people in their work in a way many models could not have previously done. They transform people’s lives so much that we cannot return to a time before Generative Applications existed.
Many famous product models were locked behind companies, so the community appreciates Hugging Face’s presence to democratize open-source Generative AI.
Many fantastic models that have changed the technical landscape are hosted in the Hugging Face, but did you know you can access it easily with HuggingChat?
This article will explore why you should use HuggingChat and understand the application’s various features.
Let’s get into it.
HuggingChat Application
HuggingChat is an application that allows users to chat using the Generative Model. If you are familiar with ChatGPT, the functionality is similar, although the UI is different. Overall, the UI looks like the image below.
Chat Features
The chat features are the same as those of the other products. You can pass the prompt you want, and the model will answer it.
However, HuggingChat generally needed access to image generation or search engine tools from the beginning. We will explore how to do that later with HuggingChat, but natively, you can only converse with the model with text input and output.
Some of the models allow us to search the web. If you see the Search Web button, you can click on it, and the chat will use the information from the Internet.
You can control the chat experience in more detail in the settings section. If you go to the bottom left part of the site, you can see the setting.
You can only control a few things from the application settings, but you can always share your feedback with HuggingFace if you want to see other Chat features within HuggingChat.
HuggingChat Models
If you are already familiar with the ChatGPT UI, you know you can change the models you use. HuggingChat provides several open-source LLM models that you can use to converse. To change the model, you can go to the settings or the left-bottom part and select the Models section.
As you can see, only 9 different models were available when this article was written. To learn more about the model, you can visit the model repository by selecting the Model Page section.
To be more precise about how the model should behave, you can go to settings and pass the System Prompt to configure it. Precisely convey how the model should act and determine which prompt works the best for your use cases.
Some models might already have their default system prompt, but you can always change it yourself.
HuggingChat Assistants
An assistant is a pre-defined model trained or set to act in a certain way. If you go to the bottom-left part, you can see the assistant section, similar to the image below.
HuggingFace and the community have developed many model assistants that we can use for our work. To select an assistant, you only need to filter or search for it. For example, I chose one of the assistants who took me to the chat UI.
You can visit settings for more details about their instructions and essential information.
You can also create your own assistant. Select the Create new assistant button, and you will see the screen below.
Fill in the information about the assistant you want and whether it can access the Internet. Press the Create button, and you will quickly have your assistant.
HuggingChat Tools
Lastly, we will discuss the HuggingChat Tools. In HuggingChat, tools are applications that the model can use to decide whether to call. To be more specific, the tools are the HuggingFace Space using Gradio, which is the input/output detected from Gradio API.
To access the HuggingChat Tools, select the Tools section at the bottom left. You will see a selection of tools like the one below.
You can see the tool name and description. If you select one of the tools, you can see the information below.
You need to activate the tools if you want the chat to call the tools. Once it is activated, the model can call that tool. However, only some models can access the tools we have activated. You can go to the model’s section and see if the model has a purple wrench icon. If it does, then you can use the model with HuggingChat Tools.
For example, I ask the model to generate an image, and they can do that by calling the Image Generation tool.
It’s also possible to develop our tools. We can provide information on creating the tool by selecting the Create new tool button, as shown in the image below.
Ensure you have access to Hugging Face Space, where your tool is hosted, and that it works well.
Conclusion
HuggingChat is a UI-based site that allows users to converse with the open-source models hosted in HuggingFace. HuggingChat has many features, including chatting with assistants and tools. It’s a versatile site, as we can host our assistants and tools for the community.
I hope that has helped!
Cornellius Yudha Wijaya is a data science assistant manager and data writer. While working full-time at Allianz Indonesia, he loves to share Python and data tips via social media and writing media. Cornellius writes on a variety of AI and machine learning topics.