"model happier and more eager to explore and innovate"
You, sir. You are the creative LLM.
Downloading for immediate testing.
Might stuff it into: https://github.com/dnhkng/GLaDOS
The GlaDOS homicidal sarcastic conversations already make me laugh, this might be a level up.
That is just... awesome :)
After the first evals I would definitely recommend Deckard :)
https://huggingface.co/nightmedia/Qwen3-Deckard-6B-qx86-hi-mlx
LinkedIn full review here:
First of all, I would like to express my respect and gratitude to you. The LIMI-Air-qx86-hi-mlx model is 97GB in size and works very well when deployed locally on a Mac with 128GB of RAM. The 4-bit quantization is too small, obviously resulting in a significant performance loss; I've also tested it, and the text processing results are not ideal. Similarly, the 6-bit quantization is not ideal either. In my personal opinion, LIMI-Air-qx86-hi-mlx is currently the most ideal model for processing Chinese text on a Mac with 128GB of RAM. The output results, in my opinion, are better than QWEN3-80b.