nicman23

i only care about the license and results to be honest

i think it is a very happy time charging whatever per hour you want

the qwen 27b is quite good for doing a mvp

you can by making a different - usually local small - model ask for verification

dense small models do not like quantization. i find 27b fp8 to be smarter albeit less knowledgable versus the 122B

to be honest i think this is a open source vs close source software.

some companies chose to have the benefits of one or the other

huh i think i can use that with a S7150 to make a cheap vdi server with the video encoding happenng in the host

tailscale is self-hostedable

but it is just a agents.md file lol

just do it!

be the change

also the counter argument is that median saas company is outsourcing to AI anyway, that being artificial intelligence or actual Indians

i really dislike that there is no way to do dhcp for new clients and that i have to manually define peers in each "exit node"

more importantly what is the cost versus getting it the non recyclable way

what the ps3 or the tinygrad?

but mostly the high prices

lol he is right. his work both on the ps3 and now with the hacked p2p drivers is powering many a lab

yeah shilling a non profit open source OS.

what is the issue with graphene?

they need to make a official GSI. until they do not they ll continue to decline.

? that is just any computer

just like full self driving tm

https://github.com/nicman23/flipper_to_IRSend

i vibed a .ir to tasmota IRSend commands and it is so good.

why it should?

gross negligence / honest mistake

if you cannot differentiate the 2, :insert rude thing here:

you understand the concept of zero days ?

companies should be better and if not, criminally liable for their bad code.

i would argue it is more valuable

lmao no.

it would be quite funny if they used the hip cuda backend to make new nvidia gpus run 32bit physx

nazi bar?