SteerLM empowers users to define all the attributes they require and integrate them into a single model.
Then, they can seamlessly select the specific combination they need for a particular use case while the model is actively running.
1. Customize an AI model predicting attribute performance using a fundamental set of prompts, responses, and desired attributes. 2. Automatically generate a dataset using this model. 3. Train the model with the dataset using standard supervised fine-tuning techniques.
SteerLM simplifies these complex and time-consuming processes into three straightforward steps:
With SteerLM, a company can develop a single chatbot that can be instantly tailored to meet customers' changing preferences, demographics, or circumstances across various vertical markets and geographic regions.
SteerLM is highly adaptable and can be applied to nearly any enterprise use case that involves text generation.
Presently, some games incorporate numerous non-playable characters who mechanically repeat predefined text, regardless of the user's actions or the situation.
NVIDIA showcased the potential of SteerLM in an unexpected area - gaming (refer to the video below).
While crafting a prototype, he recognized that a popular model-conditioning technique could be integrated into the method.
Yi Dong, an applied research scientist at NVIDIA, recalled the moment of inspiration: "I woke up early one morning with this idea, so I jumped up and wrote it down."