maybe someone will figure out some ways to prune/ quantize it a huge amount ;-;

edit: If the AI bubble pops we will be swimming in GPUs... but no new models.

This is absurd. We have crossed the point of no return, llms will forever be in our lives in one form or another, just like internet, especially with the release of these open model weights. There is no bubble, only way forward is better, efficient llms, everywhere.

You seem to not understand what a bubble popping is. Yes we have the internet around, that doesn’t mean the dot com bubble didn’t pop…

This bubble collapsing along with most blockchains going all in with proof of stake rather than proof of work is myself and every other gamer wet dream.

Your going to need a lot more than a few, 800G VRAM needed

If previous quantization results hold up, fp8 will have nearly identical performance while using 405GiB for weights, but the KV cache size will still be significant.

Too bad, too, I don't think my PC will fit 20 4090s (480GiB).

I've got a motherboard that will support 8!

40,320 4090s?? What witchcraft is this?! :D

All the more impressive when you realize that Groq's infrastructure (based on LPUs) was built using only 6!

Quantized to 4 bits you'll only need ~200GB! 5 4090s should cover it.

I wonder if AutoAWQ works out of the box, given no architectural changes (?). That would be most straightforward together with vLLM for serving.

Two 128gb Mac studios networked via thunderbolt 4?

This is actually a promising endeavor. Id love to see someone try that.

There's already at least one project that attempts this:

https://github.com/exo-explore/exo

You'll probably need 9 or more. 4090s have 24GB each.

Oops, I read 48 somewhere but that's wrong. Thanks.

a6ks however

If an implementation had NVidia's Heterogeneous Memory Management implemented, then 192 GB RAM DDR5 + GPU VRAM would seem to be close.

follow the trail of tears to my credit card

Oof.

how is this even useful? no one can run it.

You don't use the 405B parameter model at home. I have a lot of luck with 8B and 13B models on a single 3090. You can quantize them down (is that the term) which lowers precision and memory use, but still very usable... most of the time.

If you are running a commercial service that uses AI, you buy a few dozen A100s, spend a half million, and you are good for a while.

If you are running a commercial inferencing service, you spend tens of millions or get a cloud sponsor.

I can't expect all my users to have 3090s and if we're talking about spending millions there are better things to invest in than a stack of GPUs that will be obsolete in a year or three.

No, but if you are thinking about edge compute for LLMs, you quantize. Models are getting more efficient, and there are plenty of SLMs and smaller LLMs (like phi-2 or phi-3) that are plenty capable even on a tiny arm device like the current range of RPi "clones".

I have done experiments with 7B Llama3 Q8 models on a M3 MBP. They run faster than I can read, and only occasionally fall off the rails.

3B Phi-3 mini is almost instantaneous in simple responses on my MBP.

When I want longer context windows, I use a hosted service somewhere else, but if I only need 8000 tokens (99% of the time that is MORE than I need), any of my computers from the last 3 years are working just fine for it.

If you want to run the 405B model without spending thousands of dollars on dedicated hardware, you rent compute from a datacenter. Meta lists AWS, Google and Microsoft among others as cloud partners.

But also check out the 8B and 70B Llama-3.1 models which show improved benchmarks over the Llama-3 models released in April.

For sure, I don't really have a need to self host the 405b anyways. But if I did want to rent that compute we're talking $5+ /hr so you'd need to have a really good reason.

Christ!!