(dynomight.substack.com)

688 points crescit_eundo | 2 comments | 14 Nov 24 17:05 UTC | HN request time: 0.602s | source

Show context

cjbprime ◴[15 Nov 24 07:27 UTC] No.42144644[source]▶

> I ran all the open models (anything not from OpenAI, meaning anything that doesn’t start with gpt or o1) myself using Q5_K_M quantization, whatever that is.

It's just a lossy compression of all of the parameters, probably not important, right?

replies(1): >>42147420 #

1. loa_in_ ◴[15 Nov 24 14:52 UTC] No.42147420[source]▶

>>42144644 #

Probably important when competing against undecimated ones from OpenAI

replies(1): >>42150126 #

2. NiloCK ◴[15 Nov 24 19:37 UTC] No.42150126[source]▶

>>42147420 (TP) #

Notably: there were other OpenAI models that weren't quantize, but also performed poorly.

↑

Something weird is happening with LLMs and chess