mastodon
4.5.7
The views expressed here reflect the opinions of my previous employers and their spouses.
If you know me and follow me anyway and then get mad at me, that's on you
The views expressed here reflect the opinions of my previous employers and their spouses. If you know me and follow me anyway and then get mad at me, that's on you
busy fucking my way up the supply chain
The views expressed here reflect the opinions of my previous employers and their spouses. If you know me and follow me anyway and then get mad at me, that's on you
The views expressed here reflect the opinions of my previous employers and their spouses. If you know me and follow me anyway and then get mad at me, that's on you
Let's say one of the big #LLM models was exfiltrated. What sort of hardware would someone need to be able to run them? I've only seen small models being run locally and it looks like 1GB RAM per 1B parameters but idk if that's how they scale or not. So maybe around 1TB RAM to run a large commercial model? idk, just thiinking about if/when these "assets" become accessible to the people whose works they're built on and what sorts of access the public could even have given the resources needed