Releases: adammikulis/MindGame
0.2.0 - Configuration and 3D Example Update
MindGame v0.2.0 is a very large overhaul of the previous release. Model configurations can be saved and loaded at later times with all relevant parameters. I have refactored the classes extensively and there is now a dedicated MindAgent node that can be attached to any scene which refers to a MindManager autoload.
I also created a MindAgent3D (a CharacterBody3D that has an attached MindAgent), which will eventually become its own dedicated node. Overall, the usability of the addon is much better with this update and it should be easier to integrate into a project. I removed the bottom bar option for now as I am focusing on the 3D implementation of MindAgents.
The next minor update will be for inference parameters, including the ability to make all output in the format of a JSON. Autoloading configurations will be the final feature of v0.2, before moving onto Graph Networks in v0.3. Currently, LLaVa and embedding models are not implemented. LLaVa functionality will be delayed in order to transition to the BatchedExecutor, which does not support that feature.
New Features
Model Configuration
Local Chat in a 3D Scene
What's Changed
- Model Configuration, Autoload, and MindAgent node refactor by @adammikulis in #1
- Configuration and 3D Chat Example update by @adammikulis in #2
New Contributors
- @adammikulis made their first contribution in #1
Full Changelog: 0.1.3...0.2.0
Contributors
Assets 4
0.1.3 - First Release of Mind Game!
This is the first release of Mind Game, compiled for Windows and Linux with CUDA 12 support. The current plugin has functionality for loading/unloading .gguf language models and chatting with them. Future versions will include an export for Android (Mac and iOS exports not possible at this time due to lack of Apple hardware), support for image analysis (LLaVa), vector databases, and a download manager to easily obtain models.
Models:
Phi-3 Mini