I don't run large models for a living but napkin'd some maths based on provider cost sheets for existing ones and *i think* Meta's forthcoming Llama 405B is gonna cost orgs (at least) $500K+ a month to run it in any type of useful hardware+data center configuration.
You could do alot of good for humans for $500K/month instead.
boB Rudis 🇺🇦 /
npub1u9…nuwwf
2024-07-15 13:24:43
Author Public Key
npub1u99vm8p87l3n8h2wm9zs592sr28jud5t9r7xkzq4dwyze0p30j0scnuwwfPublished at
2024-07-15 13:24:43Event JSON
{
"id": "1060a5f3b8650c7d9fcb040ce9d5ee2f4de0bfcfe652e20f2a36fa3be7f5dc17",
"pubkey": "e14acd9c27f7e333dd4ed9450a15501a8f2e368b28fc6b08156b882cbc317c9f",
"created_at": 1721049883,
"kind": 1,
"tags": [
[
"proxy",
"https://mastodon.social/users/hrbrmstr/statuses/112790725155109255",
"activitypub"
]
],
"content": "I don't run large models for a living but napkin'd some maths based on provider cost sheets for existing ones and *i think* Meta's forthcoming Llama 405B is gonna cost orgs (at least) $500K+ a month to run it in any type of useful hardware+data center configuration.\n\nYou could do alot of good for humans for $500K/month instead.",
"sig": "54355387c3b54947858630c5527ab380a9e423171ff11ee6d2fa8571fa09a628bf71a4d243a340dc3c76466365818a8e8247331e5f1a3cf1c15a1c8617299716"
}