Super cool! Do you do any analysis or have any tools that help you identify these circuits? I came across this [1] recently, and wanted to try to identify specifically strong "circuits" in what seems to be a similar way to what you did.
Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
11–20 of 137 posts
Re: Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
#12Did you ever try multiple copies?
I did, but the combinatorics are mad. I have also tried training a meta-model that predicts the outputs of the combinations.
I will make another post if the topic is popular; its pretty geeky though, even more than my usual blog posts...
Re: Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
#13Here is a paper that made a similar observation recently: https://www.alphaxiv.org/abs/2512.19941
Thanks for the link!
I think that these models have to learn to efficiently use their parameters, and the best way to do that is 'evolve' (yes, a bad word for it), structures over pretraining time. Unfortunately, they don't have a way to access these structures 'from the inside'. I hope this new approach lets up boost performance in s more experimentally rigorous way
Re: Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
#14Really interesting discovery, especially the part about base64.
Reminds me of this: Transformer Layers as Painters https://arxiv.org/abs/2407.09298
Re: Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
#15Super cool! Do you do any analysis or have any tools that help you identify these circuits? I came across this [1] recently, and wanted to try to identify specifically strong "circuits" in what seems to be a similar way to what you did. [1] https://weightwatcher.ai/
I build my own analysis tools. I'm just finishing up running the current generation of LLMs (MiniMax M2.5 and the Qwen3.5 family), and then I will put it all on Github.
It less 'tool', than an assorted set of scripts, tailored to my unusual hardware setup. But it should be easy to extend; I would have released this earlier but I had the (stupid) idea to 'write a paper' on this. Aiming for that delayed this a year. Blogs are the way to go (for me).
Re: Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
#16Here is a paper that made a similar observation recently: https://www.alphaxiv.org/abs/2512.19941
Thanks for the link! I think that these models have to learn to efficiently use their parameters, and the best way to do that is 'evolve' (yes, a bad word for it), structures over pretraining time. Unfortunately, they don't have a way to access these structures 'from the inside'. I hope this new approach lets up boost performance in s more experimentally rigorous way
I think the recurrence is a consequence of using a residual connection, seems like that makes the representation stay consistent across layers
Re: Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
#17Super cool. Love seeing these writeups of hobbyists getting their hands dirty, breaking things, and then coming out on the other side of it with something interesting.
Re: Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
#18This layer duplication strikes me as a bit of "poor man's" version of looped language models: https://ouro-llm.github.io/ Pretty cool though. LLM brain surgery.
Agrees, but one thing to note:
I really think from the experiments that 'organs' (not sure what to term this), develop during massive pretraining. This also means maybe looping the entire models is actually not efficient. Maybe a better way is [linear input section -> loop 1 -> linear section -> loop 2 -> linear section -> ... -> loop n -> linear output]?
This would give 'organs' space to develop.
Re: Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
#19I am not really an ml dev so I don't understand most of it. It does sound ridiculous how it would even work work. Brilliant work and great article I enjoyed reading it This sounds similar to the Kimi's mixture of experts architecture if I understood it correctly(likely I have not), can you comment on this ?
No worries, happy to discuss anyway :)
MoE (mixture of experts), is an architecture that forces sparsity (not all 'neurons' are active during the forward pass.
This is pretty much orthogonal to that; it works with dense and MoE models, by repeating 'vertical' sections of the transformer stack.
Re: Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
#20Isn't this similar to models that have "double check the answer"?
First pass runs your input through, second pass runs it's output as input?
Just, in double check it presumably runs the entire stack while you're trying to skip the translation steps and only double check the logic?