Show HN: I built a tiny LLM to demystify how language models work • Hacker News Board

Re: Show HN: I built a tiny LLM to demystify how language models work

#81

Post by armanified » Mon, Apr 06, 2026, 11:50 AM UTC

amelius wrote: post #75

> A 9M model can't conditionally follow instructions How many parameters would you need for that?

My initial idea was to train a navigation decision model with 25M parameters for a Raspberry Pi, which, in testing, was getting about 60% of tool calls correct. IMO, it seems like around 20M parameters would be a good size for following some narrow & basic language instructions.

Re: Show HN: I built a tiny LLM to demystify how language models work

#82

Post by siva7 » Mon, Apr 06, 2026, 11:50 AM UTC

amelius wrote: post #76

_2fnr wrote: post #50

[flagged]

It's arguably even better than the most famous answer to that question.

which is?

Re: Show HN: I built a tiny LLM to demystify how language models work

#83

Post by novachen » Mon, Apr 06, 2026, 11:51 AM UTC

[dead]

Re: Show HN: I built a tiny LLM to demystify how language models work

#84

Post by armanified » Mon, Apr 06, 2026, 11:51 AM UTC

Duplicake wrote: post #72

I love this! Seems like it can't understand uppercase letters though

Uppercase letters were intentionally ignored.

Re: Show HN: I built a tiny LLM to demystify how language models work

#85

Post by armanified » Mon, Apr 06, 2026, 11:53 AM UTC

totetsu wrote: post #61

https://bbycroft.net/llm has 3d Visualization of tiny example LLM layers that do a very good job at showing what is going on ( https://news.ycombinator.com/item?id=38505211 )

Pretty neat! I'll definitely take a deeper look into this.

Re: Show HN: I built a tiny LLM to demystify how language models work

#86

Post by zkmon » Mon, Apr 06, 2026, 11:53 AM UTC

_2fnr wrote: post #50

[flagged]

Meaning/goal of life is to reproduce. Food (and everything else) is only a means to it. Reproduction is the only root goal given by nature to any life form. All resources and qualities are provided are only to help mating.

Re: Show HN: I built a tiny LLM to demystify how language models work

#87

Post by algoth1 » Mon, Apr 06, 2026, 11:54 AM UTC

This really makes me think if it would be feasible to make an llm trained exclusively on toki pona (https://en.wikipedia.org/wiki/Toki_Pona)

Re: Show HN: I built a tiny LLM to demystify how language models work

#88

Post by siva7 » Mon, Apr 06, 2026, 11:55 AM UTC

59nadir wrote: post #56

brcmthrowaway wrote: post #32

Why are there so many dead comments from new accounts?

Because despite what HN users seem to think, HN is a LLM-infested hellscape to the same degree as Reddit, if not more.

But what should we do? The parent company isn't transparent about communicating the seriousness of this problem

Re: Show HN: I built a tiny LLM to demystify how language models work

#89

Post by armanified » Mon, Apr 06, 2026, 11:57 AM UTC

nullbyte808 wrote: post #3

Adorable! Maybe a personality that speaks in emojis?

OMG! You just gave me the next idea..

Re: Show HN: I built a tiny LLM to demystify how language models work

#90

Post by armanified » Mon, Apr 06, 2026, 11:59 AM UTC

SilentM68 wrote: post #5

Would have been funny if it were called "DORY" due to memory recall issues of the fish vs LLMs similar recall issues :)

OMG! Why didn't I thought fo this first :P