HN
Hacker News
Top
New
Best
Ask
Show
Jobs
Comment by vkaufmann | Hacker News Reader
Parent
Full thread
vkaufmann
·
GPT-OSS-120B runs like hell on my DGX Spark
View on HN
embedding-shape
·
The MXFP4 variant I suppose? My setup (RTX Pro 6000) does around ~140 tok/s with llama.cpp, around 160 tok/s with vLLM.
vkaufmann
·
yep MXFP4 really fast :D
Reply on news.ycombinator.com