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

Show HN: Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training

github.com

Post reply on HN
Print view View original

51–60 of 100 posts

Re: Show HN: Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training

#51

Post by yorwba »

regularfry wrote: post #48
simgt wrote: post #46

> I replicated David Ng's RYS method [...] found something I didn't expect. > Transformers appear to have discrete "reasoning circuits" — contiguous blocks of 3-4 layers that act as indivisible cognitive units. Duplicate the right block and the model runs its reasoning pipeline twice. No weights change. No training. The model just thinks longer. How did you not expect that if you read his post? That's literally what…

That's explicitly not the unexpected part. Read the rest of the post.

After reading both the original post and this submission, what do you think is new here?

Re: Show HN: Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training

#52

Post by zozbot234 »

4bpp wrote: post #27

Assuming the benchmarks are sound (rather than capturing a fluke), the provided explanation still does not pass the smell test. As far as I can tell, there is nothing about the training process of these models that would encourage them to make the output of any layer apart from (n-1) meaningful as the input of layer n, unless perhaps these layers were initialised as identity and the training process did not get to ch…

> far as I can tell, there is nothing about the training process of these models that would encourage them to make the output of any layer apart from (n-1) meaningful as the input of layer n

Wouldn't "pass-through" identity connections have exactly that effect? These are quite common in transformer models.

Re: Show HN: Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training

#53

Post by 4bpp »

ACCount37 wrote: post #50
4bpp wrote: post #27

Assuming the benchmarks are sound (rather than capturing a fluke), the provided explanation still does not pass the smell test. As far as I can tell, there is nothing about the training process of these models that would encourage them to make the output of any layer apart from (n-1) meaningful as the input of layer n, unless perhaps these layers were initialised as identity and the training process did not get to ch…

Empirical findings tell a very different tale: all LLM layers use vaguely compatible internal representations. And middle layers in particular can be almost interchangeable - a lot of what they seems to be "iterative refinement of the same representations". Proven by various probes and ablations, but the most obvious one is probably the good old logit lens. This is likely to be shaped by tied embeddings and skips on…

That's interesting, could you point me to some source on these findings?

It seems to me that the difference between "iterative improvement" as you put it and "close to the identity" (as in the output is close to the input for most of the volume of the input space) as I put it is fairly subtle, anyway. One experiment I would like to see is what happens to the reasoning performance if rather than duplicating the selected layers, they are deleted/skipped entirely. If the layers improve reasoning by iterative improvement, this should make the performance worse; but if they contain a mechanism that degrades reasoning and is not robust against unannealed self-composition, it should make the performance similarly better.

Re: Show HN: Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training

#54

Post by 4bpp »

zozbot234 wrote: post #52
4bpp wrote: post #27

Assuming the benchmarks are sound (rather than capturing a fluke), the provided explanation still does not pass the smell test. As far as I can tell, there is nothing about the training process of these models that would encourage them to make the output of any layer apart from (n-1) meaningful as the input of layer n, unless perhaps these layers were initialised as identity and the training process did not get to ch…

> far as I can tell, there is nothing about the training process of these models that would encourage them to make the output of any layer apart from (n-1) meaningful as the input of layer n Wouldn't "pass-through" identity connections have exactly that effect? These are quite common in transformer models.

Yeah, that's what I meant with "initialised as identity and the training process did not get to change them much".

Re: Show HN: Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training

#55

Post by zozbot234 »

Imanari wrote: post #45

Fascinating! I wonder if new training techniques could emerge from this. If we say layer-1=translater, layer2-5=reasoner, layer6 retranslater, could we train small 6 layer models but evaluate their performance in a 1>n*(2-5)>6 setup to directly train towards optimal middle-layers that can be looped? You'd only have to train 6 layers but get the duplication-benefit of the middle layers for free.

Yes, training directly for a diverse mix of "looped" inference procedures makes a lot of sense as a way of allowing for increased inference-time compute. It would likely be complementary to the usual thinking approach, which essentially runs the "loop" LLM-wide - and, critically, yields interpretable output which lets us see what the LLM is thinking about.

Re: Show HN: Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training

#57

Post by BoredomIsFun »

SyzygyRhythm wrote: post #8

If running twice is good, then is running N times even better? I wonder if you could even loop until some kind of convergence, say hitting a fixed point (input equals output). I wonder if there's even a sort of bifurcation property where it sometimes loops A->A->A, but other times A->B->A, or more, rather like the logistic map fractal.

Phi-4-14b with layers duplicated (phi-4-25b) has increassed performance. Phi-4-49b has degraded vs 14b.
Post reply on HN