From e76542869200c7510abcc58383ae251e6d939d8a Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?S=C3=B8ren=20Fuglede=20J=C3=B8rgensen?= Date: Sat, 22 Jun 2024 22:47:41 +0200 Subject: [PATCH] More stuff --- index.html | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) diff --git a/index.html b/index.html index 95e9f21..b614ae3 100644 --- a/index.html +++ b/index.html @@ -22,9 +22,12 @@

How?



In particular, this "arbitrary" code could in principle be an entire LLM inference engine (Llama in our case, hence the name). -

+

It could also in principle be an entire LLM inference engine but not just in principle, relying on treating text containing magic symbols for fake "ligatures" to initialize the LLM and use it to generate text. +

+ At the end of the day, what this means is that you can just use the font to run the LLM and e.g. get text generation in any Harfbuzz-based text editor, + even if they don't come with "Copilot" like features themselves. So is this silly hack in fact a billion dollar idea?

Okay show me

Yeah, sorry, that was brief. Here's an attempt to make it make more sense:

@@ -32,6 +35,14 @@

Okay show me

Skip to 6:09 if you just want to see the font in action.

Usage

Just download llama.ttf (60 MB download, since it's based on the 15M parameter TinyStories-based model) and use it like you would any other font. Use it somewhere where Harfbuzz is used and built with Wasm support.

+

More fun with fonts

+

If this didn't seem pointless enough, here are some other weird things people have done with fonts and LLMs: +

+