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

Attention Is Off By One

evanmiller.org

Post reply on HN
Print view View original

221–230 of 347 posts

Re: Attention Is Off By One

#221

Post by cycomanic »

The I know it's on-vogue on HN to complain about academia, but the blog post is not making a good argument.

The post could have probably gotten the point across in less than 1/4 of the overall length (probably even less than 1/8th), instead the author wrapped the the post into lots of informalisms and a thinly veiled complained about academic publishing.

The result of this is reflected in the discussion here, nobody actually writes about the result/idea behind the post, instead we have ~200 comments discussing the merits of academic publishing vs blog posts and formal vs informal writing.

So I guess if you want to get your blog-post on the front page of HN it's a good writing style. If you want someone to consider and discuss the merits of your idea, maybe not so much.

Re: Attention Is Off By One

#222

Post by thomasahle »

quickthrower2 wrote: post #213
thomasahle wrote: post #189

Earlier quoted context omitted.

If you take the inner product between a lot of more or less random vectors (the key and query vectors in attention) most values are going to be close to 0. This means they contribute by e^0 to the denominator. Now, if you have a context length of say 2000, your denominator is already ~ 2000. Increasing it to 2001 doesn't really make a difference. Adding 1 to the denominator can be useful if you have softmax with just…

That simple comment is a strong counterpoint to the entire blog post? Except with the +1 denominator, it might be that the model trains all of the inputs to become very negative so softmax chucks out close to zeros, whereas it wouldn't bother before because making one prob bigger makes another smaller.

> it might be that the model trains all of the inputs to become very negative

It still can't do this because of L2 regularization / weight decay. If two vectors are norm 1, their inner product is at least -1, so with 2000 vectors that's still 2000 * e^(-1) =~ 735.

Not saying it's theoretically impossible that it could happen. But you would have to try _really_ hard to make it happen.

Re: Attention Is Off By One

#223

Post by simonster »

mlsu wrote: post #8

I don't really understand the subject matter enough, so I apologize in advance for the meta-comment... The author mentions that he would maybe have written this as a scientific paper: > I tried writing a serious-looking research paper about the bug and my proposed fix, but I lost a series of pitched battles against Pytorch and biblatex, so I figured I’d just write a blog post instead. (History is written by the winne…

It's interesting, because as a scientist who reads and writes these kinds of papers, my first impression was: This guy has a pretty big ego or is otherwise badly miscalibrated if he believes his genius idea has a "99.44%" chance of preventing outlier activations without doing any experiments.

Re: Attention Is Off By One

#224

Post by thomasahle »

Q6T46nT668w6i3m wrote: post #170
quickthrower2 wrote: post #168

Earlier quoted context omitted.

Can you elaborate? (It wouldn't be the first time there was an extraneous feature that no one has every used in some code!)

It’s useful but it’s less used than dummy tokens.

Are dummy tokens just tokens that don't have an associated input/output token? Like, a way to give more computational power to the model without splitting the text into more actual tokens?

Re: Attention Is Off By One

#225

Post by ticviking »

coldtea wrote: post #149
lamontcg wrote: post #82

Earlier quoted context omitted.

> It seems like, when you have something of substance to say, you say it. And this blog post probably could be condensed into 1/4 of its size or less with a less conversational/bloggy tone.

There are words that are added to drive the point in multiple ways, ease into it, and make the text more engaging. And there are words that are added to add empty padding, keep up academic pretenses, and appear smart. The post could have been condensed, but it would lose the former, not the latter.

Good rhetoric takes time and energy from both the author and reader

Re: Attention Is Off By One

#226

Post by simonster »

caddemon wrote: post #160
pessimizer wrote: post #114

Earlier quoted context omitted.

> The tone is self-effacing, it does not have an "ego" the way scientific papers tend to have. I can't imagine judging scientific papers based on whether the author might be looking down on me, or thinks he knows better than me. > if we were "allowed" to cite research that reads like this Maybe you're looking down on yourself? You can cite anything you want to cite.

Well if you yourself are trying to publish in a scientific venue you can't always cite exactly what you want to cite. Though it's probably uncommon for a peer reviewer to ask for a specific citation to be removed, the review process absolutely does affect the references list, and expectations about this process affect it doubly so.

In ML, no one is going to police your citation list. I've cited some weird stuff in my papers, including ideas from tweets and random quotes from Jeff Dean. It's never been a problem.

Re: Attention Is Off By One

#227

Post by gromneer »

doliveira wrote: post #30
mlsu wrote: post #8

I don't really understand the subject matter enough, so I apologize in advance for the meta-comment... The author mentions that he would maybe have written this as a scientific paper: > I tried writing a serious-looking research paper about the bug and my proposed fix, but I lost a series of pitched battles against Pytorch and biblatex, so I figured I’d just write a blog post instead. (History is written by the winne…

Nah, scientific papers are supposed to be precise and technical. This reads like those quite frequent suggestions here of switching all equations in papers to plain English or code: it honestly comes from a place of ignorance, and I say that as basically a layman myself. What should be encouraged is for academics to blog about their research as well. It would even help when recruiting and onboarding new members. Righ…

I disagree because it isn't possible for language to be precise on it's own syntactic merit. There is meaning and there is context and the biggest problem with research papers is that the context of many statements in the paper are incredibly ambiguous. The reason for that is that the papers are trying to be "concise". Context can only be disambiguated with more statements. You must eliminate potential interpretations that a reader could make.

"Spectrum sharing in an “apple-like” or a fixed set sense is not a coexistence. ". What does that mean? Coexist? Who knows, the author thought they were being precise, but they understood the statement they made with a head full of context that gave it precise meaning. As readers, we can only scratch our own heads as to what that context could possibly be.

Re: Attention Is Off By One

#228

Post by lostmsu »

sbszllr wrote: post #63
lostmsu wrote: post #51

Earlier quoted context omitted.

What's wrong with unscaled matrix multiplication? Softmax has some kind of intuition in the context, but why not layer norm or something else instead (if anything is needed at all)?

The family of sigmoid functions has nice gradient properties with theoretical backing. Good starting read: https://stats.stackexchange.com/questions/162988/why-sigmoid...

Doesn't really explain why not more typical ReLU + layer norm or some alternative.

Re: Attention Is Off By One

#229

Post by lyjackal »

chessgecko wrote: post #152
zackangelo wrote: post #136

Earlier quoted context omitted.

He’s not trying or claiming to improve attention. He’s trying to reduce outliers to improve the ability to quantize the parameters.

He refers all over the blog post to an "error" in attention. specifically says The problem with using softmax is that it forces each attention head to make an annotation, even if it has no information to add to the output vector. Using softmax to choose among discrete alternatives is great; using it for optional annotation (i.e. as input into addition) is, like, not cool, man. I'm saying it uses the current position…

I agree that he used the term error somewhat incorrectly. But he seems mainly to just be making the point that sumac introduces a large outlier which in turn is only now an issue that the community is now aggressively trying to quantize models

Re: Attention Is Off By One

#230

Post by gromneer »

rTX5CMRXIfFG wrote: post #211
karaterobot wrote: post #154

Earlier quoted context omitted.

The writing quality of academic papers is very poor, whatever its intended characteristics are, and we deserve better. I'm skeptical that the only way for them to be precise and technical is to make them impenetrable. I think there is a culture of academic writing (many different cultures, really) that has adopted a voice and writing style which became a parody of itself over time. Here's a trivial example: You frequ…

Either your example is too trivial to justify your point, or the point itself is trivial. It's right for an academic to distance themselves from the subject of their study because we do need researchers who try not to be biased. If they fail that and then correct themselves, then what's the problem? Complaining about inconsequential uses of tone is obsessing about form over function and reeks too much of insecurity,…

They aren't magically "objective" because they used the passive voice. It's a performance.
Post reply on HN