luna
luna reposted
luna reposted
luna reposted
for those that may be interested, two of our story telling models are open weights now, ggufs included! huggingface.co/NovelAI/krak... (gpt-neox-20b finetune) huggingface.co/NovelAI/clio... (finetuned on a private stablelm 3b pretrain)
we have pressed the "release base/pretrain model" button internally huggingface.co/NovelAI/nove...
"running vllm at scale" is an NP-hard problem
also we we've been working for months to reproduce deepseek quality of inference. it is difficult like $100M budget + right connections + getting the right 10 people in the world to help you difficult if you ever see anyone claiming they can do this, they are lying
the USA cant fund dictatorships in LATAM anymore so china will be giving us the promise of Cool Infrastructure instead. this says something about society, idk what yet
Xi Jinping calls for a "consensus-based global AI governance framework" and says China will pioneer the establishment of a BRICS AI open-source community (Joanna Ossinger/CNBC) Main Link | Techmeme Permalink
if i follow you then you have the right to ask for @nagare.pds.bsky.ln4.net access so you too can see what happens when i get to design a general-purpose feed
i could use gh pages or i could use the cool sexy @wisp.place provenance notice: im blinking once to show i was not threatened to make this post and i do this out of my own volition
luna reposted
ive put the ~weeks of looking&writing on harnesses into an article pi/omp/polytoken/maki