Hacker Newsnew | past | comments | ask | show | jobs | submit | dackdel's commentslogin

for some reason i read that in archer(animated) voice.

Probably because it's a dumb useless comment in the same vein as most of that show.

I read this in a whiny high pitched voice with my nose and lower lip pulled up.

what kind of hardware do you need in order to run qwen3.6-27b

Depends on which variant you pull down, but a single 5090 GPU (I know these are insanely expensive, but for context) could run either the Q8 or Q4_K_M version. It will not fit the 52GB version (BF16) on the other hand. So any modern Mac with a Pro or better processor and more than 52GB of RAM (don't forget VRAM for context window also matters!) would suffice, as someone else noted, probably a 128GB model would do the trick, and give you enough wiggle room to max out the context window.

My Mac only has 16GB of VRAM (20GB total - 8 is reserved for the OS) so I have to leave room for VRAM, I usually find a model that fits in 5 to 7 GB of VRAM and then max the context window as much as I can.


The benefit of running the full precision version is negligible (probably not even measurable above the benchmark noise floor). Most common for cost-conscious users is to run something around 4-6 bits per weight, which would fit on a 24 or 32 GB card (as you mentioned).

Note you can change the amount of shared (V)RAM reserved for the OS with:

sudo sysctl iogpu.wired_limit_mb=18800

will allow you to use more, but you do need to leave a bit for the OS obviously!


Oh man! I had no idea I could do this at all! What do you usually tweak it to? I feel like 8 GB is probably still a reasonable amount to give the rest of the OS.

I've got a 32 GB MBPro, and I set it to 27700, which I haven't seen a problem with so far.

Makes sense, in my case, I've got 24GBso I guess, cranking mine to roughly 20 might not hurt?

thanks

I recommend MacBook M5 Max with 128 GB of RAM to run it comfortably and fast. If you have something like a regular M4, go with qwen3.6-35b-a3d - the Mixture of Expert architecture makes it run 2-3x faster than the 27b version.

thanks

I could run it on 7900 XT with 64k context. You could run it more comfortably on a 24 gb vram.

thanks

I bought r9700 for about 1700-1800$ and I have like 800t/s prompt and about 50t/s of inference on average? It hurt a bit when you change a prompt so llama.cpp have to discard entire cache and it have to think for 2-5min depending on the context, but otherwise it is faster than I can read.

govts are dumb

id pay to watch mkbhd(or similar) review the apc-2. and compare one made on apc-2 to someone like recordcut(or similar). that said. im glad companies like teenage are catering to the whimsy. because why not. im sure it will sell out. and they will stop producing it after they have scratched the itch of wanting to create a product like it. and hopefully after that we can get our hands on the un-redacted files.

Why would you pay for a review from a smartphone reviewer who likely has never listened to Vinyl?

Shill fetish

[flagged]


"id pay to watch mkbhd" to "fuck that smartphone reviewer guy, dont give a shit about him and hope he gets hit by a bus"

That escalated quickly... Are you ok?


trying? its like saying israel is trying to bomb iran. cars ARE spying on you.


When in school and we learn bits of history, (mostly day dreaming but sometimes information crept in) things like Shah Jahan cutting off all the hands of the sculptors of the taj mahal. I really wish Steve was alive and took inspiration, so that Jony wouldn't create trash like this.


Either you were still day dreaming, or your school history class was pretty bad. That Taj Mahal story is a myth.


I definitely think they could have made it more sporty, and that might have hit a sweet spot. Personally I love it, and that extreme difference in opinion is exactly why I think it'll be iconic. Also I wonder if you've earned the harsh criticism you spew. I doubt it.


This piqued my interest but I learned this is actually a myth.


Might be inspired by the Kremlin building. Same story but with Ivan and eyes.


that one's also a myth


thats the one thing we all have in common. we all die. that said. everything in moderation and definitely avoid a few things like sugar and maybe seed oils. but butter red meat and cheese, rather be dead.


a fully disconnected car that does not report back to its mother ship. does. not. exist. only other option is to buy a car old enough that does not have it. also if you didn't bring this up most north americans would be blissfully unaware, as long as the car has a good cup holder.


'course it does .. any custom build shop will leave such things out on request, a great many don't even add in remote networking to begin with.

eg: https://www.okaaustralia.com/


that's illegal in the EU, the car is mandated to be able to call 112 automatically, therefore it must have a cellphone in it


This can be a completely independent unit. In fact, with all the safety-related certifications I bet that's even the easiest and cheapest way to do it!


Custom builds are exempt from this requirement.


I swear I didn't know the antennae of the tracker ~~safety~~ device was wrapped in aluminum all this time!


This is the one feature I'd actually like to have. It's a shame the adnet has to abuse everything.


> as long as the car has a good cup holder.

The lack of a cup holder is the only thing I would change about my '98 Toyota MR2


https://zoo.dev/ allows you to re-iterate on the same model over and over with prompts, without resorting to creating a new model from scratch every time.


I'm pretty sure they took that from us. We had the first conversational ux in the space.


Just found this. Reminded me of Ente (which is just for photos). Has anyone used parachute? I am just curious.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: