Linking HumanFirst Studio With A Fine-Tuned LLM
This article steps through a practical example of performing NLU Design using HumanFirst Studio. Creating a custom fine-tuned LLM model and leveraging the model within the HumanFirst Studio for NLU Design.
I’m currently the Chief Evangelist @ HumanFirst. I explore and write about all things at the intersection of AI and language; ranging from LLMs, Chatbots, Voicebots, Development Frameworks, Data-Centric latent spaces and more.
As the use of LLMs increase the need for creating custom fine-tuned LLMs will also grow in demand. And this will necessitate an ongoing process of model-maintenance, if you like.
I wanted to create a set of four articles that covers the complete life cycle of a custom fine-tuned LLM.
As you can see in the image below, I kicked this process off by converting unstructured data into NLU Design data via the HumanFirst Studio. The data was subsequently used to fine-tune the LLM.
I performed some basic tests on the LLM and compared the results to the HumanFirst NLU.
1️⃣ How To Fine-Tune A Large Language Model Using HumanFirst & Cohere
2️⃣ Cohere Fine-Tuned Large Language Models & Intent Recognition
3️⃣ HumanFirst & Cohere QuickStart Guide
This article will close the loop, by linking the custom fine-tuned LLM to the HumanFirst Studio. This will illustrate how Studio can be used to leverage a LLM in a no-code fashion for accelerated and intelligent NLU Design.
Linking HumanFirst Studio With A Fine-Tuned Cohere Model
I followed the basic process outlined here to connect my HumanFirst Studio Workspace to the Cohere LLM.
The only difference is, in this instance a fine-tuned model instead of one of the standard models are accessed.
Custom models go to sleep if not used. As seen in the image below, if the model is Asleep, the Wake button must be clicked. The model status changes from Asleep, to Training. Once the model status is Ready, the Cohere model can be added to the HumanFirst Studio workspace.
The model ID is used when adding the Cohere custom model to the HumanFirst Studio.
And lastly, in the HumanFirst Workspace Settings, under Default latent space, the Cohere custom fine-tuned model shows up for selection.
From here relevant data can be imported for processing by leveraging the custom fine-tuned LLM.
I’m currently the Chief Evangelist @ HumanFirst. I explore and write about all things at the intersection of AI and language; ranging from LLMs, Chatbots, Voicebots, Development Frameworks, Data-Centric latent spaces and more.
