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

381–390 of 435 posts

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

#381

Post by oarsinsync »

diggan wrote: post #332
lokimedes wrote: post #284

We are lucky that Alibaba, Meta and Mistral sees some strategic value in public releases. If we it was just one of them, it would be a fragile situation for downstream startups. And they’re even situated in three different countries.

> We are lucky that Alibaba, Meta and Mistral sees some strategic value in public releases. Now if we only can get Meta to understand what "Open Source" means so the word doesn't lose all meaning in the future.

> so the word doesn't lose all meaning in the future

The English language doesn’t have a central body that decides what words get to exist, and what they mean, like e.g. the French.

English dictionaries are descriptive, not prescriptive. The definitions are defined by popular culture, rather than a central body.

What is needed is a large enough mass of people to recognise a word as having a particular definition, in order to change/add that definition. See “literally” having a definition of “for emphasis, not literally” as a prime example of the wisdom of crowds.

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

#382

Post by cma »

imdsm wrote: post #316
echelon wrote: post #315

Earlier quoted context omitted.

Most of the papers in machine learning are coming from China. The vast majority. Most of the open source models on GitHub, too. Hailuo, Kling, Vidu, and Hunyuan (posted on Banodoko) blow Sora and Runway out of the water. China is dominating at this field. And if they begin releasing weights as open source, that'll mean foundation model companies can only bank on the thin facade of product. That's a really good strate…

Are they making breakthroughs or are they taking what exists and copying/marginally improving?

I think they had the earliest paper with the modern take on scaling laws, from Baidu about 3 years before the similar paper from the US.

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

#383

Post by skyyler »

JKCalhoun wrote: post #372
stavros wrote: post #327

Earlier quoted context omitted.

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.

>other clueless readers that are learning

Indeed! I don't pay attention to this space often and this chain of comments has been enlightening. Thank you for being vocally clueless while we remained mostly-silently-clueless, JKCalhoun! :-)

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

#384

Post by ponector »

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 are right. I've been looking at MacBook with M4.

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

#385

Post by sangnoir »

pton_xd wrote: post #367
delusional wrote: post #343

Earlier quoted context omitted.

For the "How many r's in the word strawberry" total duration: 15.278476756s load duration: 14.982999ms prompt eval count: 47 token(s) prompt eval duration: 5ms prompt eval rate: 9400.00 tokens/s eval count: 377 token(s) eval duration: 15.257s eval rate: 24.71 tokens/s

15 seconds to count the number of r's in a string using 120 TFlops... Really puts into perspective that we've got a long way yet to go.

This specific model is designed to second-guess itself/ chack it's work. So the slowness is part of the point.

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

#386

Post by dleary »

sangnoir wrote: post #363
api wrote: post #360

Earlier quoted context omitted.

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-…

> To recapture the spirit of open source as being about freedom for actual users (as opposed to free labor for jailed SaaS) That spirit was never there; "Open Source" was created to be corporate -friendly as it was predated by Free software, which is rigidly committed to users freedom.

That is a pretty cynical take. FSF good, OSI bad.

> ”Open Source" was created to be corporate-friendly as it was predated by Free software, which is rigidly committed to users freedom.

Rigidly committed to a certain interpretation of users freedom. And that interpretation happens to involve removing a number of freedoms “for your own good”.

It is more correct to say that the GPL was created to be anti-corporate, as it was predated by both the MIT and BSD licenses, which are more free, both for users and corporations, which the FSF finds intolerable.

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

#387

Post by seanmcdirmid »

ponector wrote: post #384
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 are right. I've been looking at MacBook with M4.

You would need to get an M4 Max (which is probably the right choice for real LLM work). I don't think mobile GPUs do very well on the PC side, so if you want portability, Apple might be your best bet.

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

#388

Post by dleary »

denzil wrote: post #271
xanderlewis wrote: post #263

Earlier quoted context omitted.

It seems obvious to me that she has one sister. Or is that the naive, wrong answer?

While Sally is usually girl's name, the question never states that. So Sally could be actually a boy and in that case Sally would have two sisters.

You’ll get some eye rolls from people when you bring that up for Sally.

But there are several names that used to be considered male but are now female, like Leslie and Marion. I don’t think I’ve ever met a man name Marion, but you still occasionally run into a Leslie.

It would be interesting to start using Leslie for this little logic puzzle and see how that affects people’s answers.

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

#389

Post by int_19h »

Datagenerator wrote: post #258
int_19h wrote: post #164

Earlier quoted context omitted.

Since this is a local model, you can trivially force it to do pretty much whatever you want by forcing the response to start with "Yes, sir!".

Any prompt or system setup examples which work well?

There's no need to tweak the default prompt with this approach. Just make sure that, at the point when the model starts generating, it already has "Yes sir!" as the first tokens of the response message.

It's very easy in the API, obviously, but most local chatbot apps can also do this. E.g. in text-generation-webui, there's literally a textbox in the chat labelled "Start reply with". In LM Studio, you can pre-create a response message with the desired prefix and then use the "Continue" action on it.

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

#390

Post by sebastiennight »

Lucasoato wrote: post #35

> Find the least odd prime factor of 2019^8+1 God that's absurd. The mathematical skills involved on that reasoning are very advanced; the whole process is a bit long but that's impressive for a model that can potentially be self-hosted.

Wait I didn't read the source of that quote, but is the answer "2"?

Since (2019^8 (an odd number) plus one) is an even number, it only has one prime factor which isn't odd : 2.

Post reply on HN