Dispersion loss counteracts embedding condensation in small language models

21 pointsposted 3 hours ago
by E-Reverance

5 Comments

aetherspawn

2 hours ago

It makes sense to me that distributing across more parameters results in models that can be quant more heavily (information theory - more bits available)

I wonder if anyone has figured out how the information is compressed and calculated the amount of information an LLM can hold depending on its size

woadwarrior01

an hour ago

> I wonder if anyone has figured out how the information is compressed and calculated the amount of information an LLM can hold depending on its size

You might want to look at Physics of Language Models[1]. IIRC, the authors estimate it to be ~2 bits of factual knowledge per parameter.

[1]: https://physics.allen-zhu.com/

lwansbrough

2 hours ago

Anyone with a billion dollars want to try this and report back?

nullc

2 hours ago

From the paper it appears that it's probably more useful on small-ish models.

lwansbrough

44 minutes ago

What does it cost to train a model like 1-bit Bonsai? Anyone know?