Hacker News
new
past
comments
ask
show
jobs
points
by
mirekrusin
22 hours ago
|
comments
by
jakswa
21 hours ago
|
[-]
some support already merged, and I verified in a local build that it runs (cannot get MTP params working tho, about ~40 tok/s on my beefy 800GB/s 7900XT w/ 20GB VRAM).
https://github.com/ggml-org/llama.cpp/pull/26841
reply
by
bwfan123
20 hours ago
|
parent
|
[-]
Just tested muse-glimmer:30b-mlx on my laptop. Works great although a bit slow.
reply