Will someone figure out how to train decent Stable diffusion models for <$10K USD? (2024 end)
AITechnical AI TimelinesAI Impacts
Basic
13
Ṁ144
Dec 31
63%
chance

The original price was 300k (mid 2022)

Mosaic slashed it down to 50k by using groupnorm and layernorm with reduced numerical precision as well as a bunch of other optimizations and tweaks.

Get Ṁ1,000 play money
2 Comments
9 Holders
16 Trades
Sort by:

can you specify a bit on "decent" :) I.e. I suspect it's very possible to achieve 90% of the quality by good curation of the dataset.

@PeterBuyukliev I expect great improvements by dataset curation being high quality as well. If it leads to such incredible cost reductions, then why not?

"Decent" - sigh, it'll always be a bit subjective, but I'll be reasonable about it.

Related questions