[XRCLOUD] Try connecting Yanolja AI and LLAMA3 in the WebXR Metaverse
Hello, this is Luke from BELIVVR. I’ve shared that XRCLOUD is preparing to launch new assets as InlineViews that are easily extensable with web services. Recently, Meta has released AI technology called LLAMA3 that can be run on personal PCs. The Korean accommodation IT corperation, Yanolja, has also released a Korean AI custom model, EEVE-Korean, which now allows anyone to easily upload and use AI models.
In this regard, I will quickly test connecting this AI model and introduce the method.
First, I went to XRCLOUD’s Room menu and downloaded an animation from Sketchfab in the Spoke space, featuring a character with an impressive hairstyle, Miss Sense of Frieren: Beyond Journey’s End series.
I have uploaded a transparent PNG for the InlineView Asset, and linked it to ollama, which can run AI models. I haven’t hosted it yet, but tested and captured it on my PC, which is equipped with an RTX 3090. I first uploaded the Korean model, EEVE-Korean, and created a page in a prompt format. Unfortunately, the AI did not answer the questions perfectly as it was not customized to the character.
Yanolja AI EEVE-Korean Model and XRCLOUD (Metaverse) Integration Video True to the Korean model, it answers well in Korean. This was my first time using LangChain, and I couldn’t demonstrate it using the chat type instead of prompt type due to a network error.
This time I uploaded the LLAMA3:8B model.
As expected, it doesn’t answer well in Korean. So, I asked, “Do you know kimchi, Do you know Gangnam Style?” and it responded well. It’s amazing that such AI can be run locally.
Now, using this, it seems possible to implement features like triggering animations in the room when certain phrases are spoken by customizing the AI to characters. The synergy between AI and the Metaverse is just beginning.
If you want to experiment with AI in the Metaverse, please visit XRCLOUD
#XRCLOUD #Metaverse #AI #ollma #LLAMA3 #MozillaHubs
