Will a single model running on a single consumer GPU (<1.5k 2020 USD) outperform GPT-3 175B on all benchmarks in the original paper by 2025?

Ṁ1kṀ4k

resolved Jan 6

Resolved

YES

ALL

There are no restrictions on the amount or kind of compute used to *train* the model. Question is about whether it will actually be done, not whether it will be possible in theory. If I judge the model to really be many specific models stuck together to look like one general model it will not count.

Market context

Technical AI Timelines

Get

1,000

to start trading!

🏅 Top traders

#	Trader	Total profit
1		Ṁ40
2		Ṁ39
3		Ṁ19
4		Ṁ15
5		Ṁ14

People are also trading

How fast will you be able to train a GPT-2-level AI on a consumer GPU in 2030?

609

Will a language model that runs locally on a consumer cellphone beat GPT4 by EOY 2026?

83% chance

In what year will a GPT4-equivalent model be able to run on consumer hardware?

2026

In what year will a GPT4-equivalent model be able to run on consumer hardware?

2026

Will a GPT-3 quality model be trained for under $1,000 by 2030?

90% chance

Will a GPT-3 quality model be trained for under $10.000 by 2030?

99% chance

Will $10,000 worth of AI hardware be able to train a GPT-3 equivalent model in under 1 hour, by EOY 2027?

16% chance

Will a GPT-4 quality model be trained for under $10.000 by 2030?

90% chance

Will it be possible to disentangle most of the features learned by a model comparable to GPT-3 this decade? (1k subsidy)

58% chance

Before 2028, will anyone train a GPT-4-level model in a minute?

Sort by:

I can't easily point to a single model that does this since no one publishes their training curves, but given how close we were two years ago and how good very cheap models are relative to GPT-3, I feel comfortable resolving this YES.

Llamas on pixel 7s https://github.com/rupeshs/alpaca.cpp/tree/linux-android-build-support (ik ik its not over 13B yet, just sharing progress)

predictedYES

There are people who run 30B Llama on consumer PC successfully and even 65B (but it is extremely slow)

@ValeryCherepanov By "run on a single GPU" I mean the weights + one full input vector can fit on a consumer GPU at once. Otherwise the question would be meaningless - you can always split up matrices into smaller blocks and run the computation sequentially.

This is now extremely close to being resolved by Llama (Llama 13B does not actually beat GPT-3 on every measured benchmark, however, it only comes very close). 72% is way too low though so I guess whoever reads this comment first can collect some free mana in expectation.