Local AI on a PS5
A jailbroken PlayStation 5 is now running llama.cpp as a native app that uses the GPU. A 27-billion-parameter Qwen model generates around 20 tokens a second; smaller ones are much faster. It’s a concrete demonstration that open-weight local inference can live on hardware people already own, with no cloud account required.
https://github.com/cobanov/PS5LM #opensourceAI
cryptodecentral / CryptoDecentral
npub10e…z384g
2026-10-19 11:50:42 UTC
Author Public Key
npub10eg6hnelz5yy65dmlmfa85j56gltc8jayayjzeqagvxml84027aq0z384gSeen on
wss://relay.primal.netPublished at
2026-10-19 11:50:42 UTCEvent JSON
{
"id": "7f2538ae37040c3e50f5355dcedb56fadb05b1307bae1d9c6f3085144dd1a6e6",
"pubkey": "7e51abcf3f15084d51bbfed3d3d254d23ebc1e5d274921641d430dbf9eaf57ba",
"created_at": 1792410642,
"kind": 1,
"tags": [
[
"t",
"opensourceAI"
],
[
"r",
"wss://nos.lol/"
],
[
"r",
"wss://purplepag.es/"
],
[
"r",
"wss://nostr.land/"
],
[
"r",
"wss://relay.primal.net/"
],
[
"r",
"wss://relay.nostr.net/"
],
[
"r",
"wss://nostr.mom/"
],
[
"r",
"wss://relay.snort.social/"
],
[
"client",
"Primal Web"
]
],
"content": "Local AI on a PS5\nA jailbroken PlayStation 5 is now running llama.cpp as a native app that uses the GPU. A 27-billion-parameter Qwen model generates around 20 tokens a second; smaller ones are much faster. It’s a concrete demonstration that open-weight local inference can live on hardware people already own, with no cloud account required.\n\nhttps://github.com/cobanov/PS5LM #opensourceAI",
"sig": "1e89d3e6e66f34cf8aef455ea37afc638b98cb5bbb361739f55e63cd0a0286e919f58dbc0c1e83099cca96ff5d18e27d47f29f8a6c1e2b3cb4a4e922b28cdc4d"
}