Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

QwQ: Alibaba's O1-like reasoning LLM

qwenlm.github.io

Post reply on HN
Print view View original

371–380 of 435 posts

Re: QwQ: Alibaba's O1-like reasoning LLM

#371

Post by kaba0 »

apocadam wrote: post #225
exact_string wrote: post #214

Earlier quoted context omitted.

These tests always make me wonder: What qualifies as a valid pattern rule? For example, why wouldn't "0" be a correct answer here (rule being "every other number on the right should be 0, other numbers do not have a pattern")?

Exactly, it's completely arbitrary. I like to frame it in terms of fitting n points (the existing data in the problem) to a polynomial of degree n+1, where there's an infinite number of ways to pick a solution and still satisfy the initial data.

Maybe the "solution with the lowest Kolmogorov complexity".

In a sibling comment, I replied that usually a repeating pattern can also be applied, but that one usually requires storing the n-sequence, rarely making it the shortest encodable rule.

Re: QwQ: Alibaba's O1-like reasoning LLM

#372

Post by JKCalhoun »

stavros wrote: post #327
JKCalhoun wrote: post #298

Earlier quoted context omitted.

Thanks. You're right, I am a layman. I may also have been focusing on the "open" LLMs since they seem to get the most talked about on HN.

What do you mean by "open"? AFAIK Qwen, Deepseek etc are just as open as the rest.

I'm learning quite a bit today.

Hopefully there are some other clueless readers that are learning from my dumb comments.

Re: QwQ: Alibaba's O1-like reasoning LLM

#373

Post by alwayslikethis »

seanmcdirmid wrote: post #366
ponector wrote: post #362

Earlier quoted context omitted.

>>not that expensive to get an M4 pro with 64GB $3900 as the starting price does not sound 'not that expensive' for me.

M4 Pro with 64GB of ram starts at $3900? It doesn't take that much searching to know that is BS: https://www.apple.com/shop/buy-mac/mac-mini/apple-m4-pro-chi... An upgraded M4 Pro mac mini to 64GB is $1999. You'll probably want the extra GPUs, though, bringing you up to $2199. But you probably want the max (not offered for the mini) or ultra to do serious LLM work. An M4 Max in a 14 inch MBP with 64GB of ram will cos…

You're also going to want some more SSD space, because working with these model files on a 512 GB root is going to quickly become a problem. If you get the 2TB option, that's another $600.

Re: QwQ: Alibaba's O1-like reasoning LLM

#374

Post by alwayslikethis »

do_not_redeem wrote: post #346
Svoka wrote: post #329

Earlier quoted context omitted.

what do you mean? I can easily run 70b on my macbook. Fits easily.

Yes and I can fit 70b in system RAM on my linux box. But parent was talking about discrete GPUs, and shared memory architectures like Intel's integrated GPUs or Apple's unified memory will never perform as well as dedicated GPU memory.

It is slower, but I would say fairly usable for just LLM inference (still much faster than you can read). I'd say it's more of a matter of control. You can be reasonably sure that your linux box is free from ordinary forms of telemetry, which is hard to do with a Mac.

Re: QwQ: Alibaba's O1-like reasoning LLM

#375

Post by jazzyjackson »

api wrote: post #360
immibis wrote: post #347

Earlier quoted context omitted.

Words mean whatever's convenient to the bottom line, which is why the OSI (a consortium of Amazon, Google, Microsoft etc) still doesn't recognize the SSPL, as it would be particularly inconvenient for clouds.

So you're saying open source doesn't exist to be free labor for SaaS? The OSI is fully captured by companies with a vested interest in promoting that model and/or using open source to 'dump' on the market and commoditize their compliments. To recapture the spirit of open source as being about freedom for actual users (as opposed to free labor for jailed SaaS) and a mutualistic gift culture (as opposed to a take-take-…

compliment vs complement :^)

Re: QwQ: Alibaba's O1-like reasoning LLM

#376

Post by sebastiennight »

bofadeez wrote: post #345

It still fails at very simple stuff. E.g. "I put an ordinary rock into a glass of water. I then turn the glass of water upside down, do a little dance, and then turn the glass right side up again. Where is the rock now?" 100+ lines later... "The rock is at the bottom of the glass, submerged in the water." Models from a year ago get this right sometimes https://pastebin.com/em5TT4Zn

Llama3.1 8B fails as well.

After I call the model on it, my favorite part of the "fixed" answer is step 3:

> Bien sûr ! Let me retrace the steps with you.

    01. You put an ordinary rock into a glass of water.
    02. Then, you turn the glass upside down.
    03. At this point, the rock would be suspended in the air above the glass, as it can’t fall out (it’s not being held by anything).
    04. Next, you do a little dance. This doesn’t affect the rock’s position.
    03. Finally, you turn the glass right side up again.
> So, when you turn the glass right side up, the rock would simply fall back into the glass, ending up at the bottom where it started.

Re: QwQ: Alibaba's O1-like reasoning LLM

#377

Post by seanmcdirmid »

alwayslikethis wrote: post #373
seanmcdirmid wrote: post #366

Earlier quoted context omitted.

M4 Pro with 64GB of ram starts at $3900? It doesn't take that much searching to know that is BS: https://www.apple.com/shop/buy-mac/mac-mini/apple-m4-pro-chi... An upgraded M4 Pro mac mini to 64GB is $1999. You'll probably want the extra GPUs, though, bringing you up to $2199. But you probably want the max (not offered for the mini) or ultra to do serious LLM work. An M4 Max in a 14 inch MBP with 64GB of ram will cos…

You're also going to want some more SSD space, because working with these model files on a 512 GB root is going to quickly become a problem. If you get the 2TB option, that's another $600.

Thunderbolt 5 mostly takes care of the storage problem.

Re: QwQ: Alibaba's O1-like reasoning LLM

#378

Post by sangnoir »

lokimedes wrote: post #286
JKCalhoun wrote: post #285

Is Alibaba's LLM the "Chinese LLM"? It would appear to have been a U.S.-only game until now. As Eric Schmidt said in the YouTube lecture (that keeps getting pulled down), LLM's have been a rich-companies game.

Deepmind is in the UK and Mistral is in France? Alibaba has been pumping out a bunch of useful models for a long time.

It is officially "Google Deepmind" now, according to their website - https://deepmind.google

Re: QwQ: Alibaba's O1-like reasoning LLM

#379

Post by sangnoir »

ddtaylor wrote: post #299
JKCalhoun wrote: post #285

Is Alibaba's LLM the "Chinese LLM"? It would appear to have been a U.S.-only game until now. As Eric Schmidt said in the YouTube lecture (that keeps getting pulled down), LLM's have been a rich-companies game.

What lecture is this?

The a 2024 lecture with Stanford students where he also insinuated that Google lost its edge because it coddles its engineers too much.

Re: QwQ: Alibaba's O1-like reasoning LLM

#380

Post by cma »

sangnoir wrote: post #379
ddtaylor wrote: post #299

Earlier quoted context omitted.

What lecture is this?

The a 2024 lecture with Stanford students where he also insinuated that Google lost its edge because it coddles its engineers too much.

In a recent interview he said where he was saying they are 2-4 years behind, with these o1-like releases he thinks he was wrong on that and they are now only one year behind. And I think that was after the earlier non-open one from China and not knowing about this one yet. I've heard mixed things on whether this one is even more capable than that earlier one from a week or two ago, but sounds like it is pretty close.
Post reply on HN