Works phenomenally 🚀
#1
by McG-221 - opened
After loading with LM Studio, only about 44.5 GB in unified memory... basically a healthy amount of space left for KV cache.
The smarts I specifically was looking for are intact! That's a win, as this MLX quant is around 42% faster than the unsloth GGUF I used previously 🙌
Thanks again, much appreciated!