Hi there,
I couldn’t find any document about the *.scene and *.prefab structure and properties.
I want to modify the files externally to check if I can do some integration with AI such as GPT or other LLMs.
Any help would be great!
Unfortunately, there is currently no documentation available for this type of information. But we will include that in our plans. If you have any questions, please feel free to ask in the forum and we will answer your questions.
I need a full description of these 2 files to train a model for making a declarative language model.
My purpose is to make a simple ai model to make UI/UX just by explaining the needs.
Dear Cocos developers,
As you may know, Meta released Llama 3 LLM days ago, I’ve researched this LLM, and I can say it is a cutting-edge open-source model and is a game changer.
I’m trying to create agents to automate some game-creating processes, but due to a lack of documentation I can’t go more for UI aspect!
This piece of content is not open sourced
Oh, I need to take it by myself though!
Is there any plan to make it open source, I meant especially this part, not the others?
Just create scenes and prefabs with nodes and components, and store the file, then open them in vscode.
I believe you can summarize the patterns through the information in the file easily and quickly.
It’s just a simple JSON structure.
Yes, I see, I know it is a simple JSON, but the catch is I don’t have any clue what some properties are! Properties like: persistent, _objFlags, data!
Importantly, It seems to be hard to integrate JSON files with LLMs!
I have to do more R&D even maybe trial and error!
Yep, because the file format of the scenes and prefabs will be changed to match the engine’s needs when required, and developers don’t need to know the details, the engine will handle all changes to ensure it’s always compatible.
So I don’t recommend deep digging into the file format. If you have special needs, you should go through it by yourself.
ahahaha. good luck. I bet you can do it.
Thanks, I’m thinking of making a new file format that covers all that is needed for integrating with LLMs and then converting it to cc standard!
It is a challenging task, but it is worth trying.
As one of the cocos developers I am looking forward to your work
1 Like
To be honest, I’ve created a small dataset for cocos creator, cocos reactive, and our framework for “Chat With RTX” and I’m getting responses with 60~70% accuracy, by using Llama 2, I need it to be more accurate.
However, this is a very early step but I believe it is a big step forward.
I’m thinking of a Devin-style application for game development, I know it is ambitious but will try!
Can I experience it?
You can download “Chat With RTX” here, however, for better experiences you need a 40* nvidia graphic card.
I’ve tested on RTX 4090