Skip to content
Better HN
Top
New
Best
Ask
Show
Jobs
Search
⌘K
undefined | Better HN
0 points
tommy_axle
24d ago
0 comments
Share
Pick a decent quant (4-6KM) then use llama-fit-params and try it yourself to see if it's giving you what you need.
0 comments
default
newest
oldest
gunalx
24d ago
I habe found llama-fit sometimes just selects a way to conservative load with VRAM to spare.
j
/
k
navigate · click thread line to collapse