It's gonna be worth it. I will be able to buy a mi355x rack for 2k dollars soon

I don’t love them, but they are soooo practical specially for beans

Not sure what kind of beans you get locally, but here we almost always cook them with a pressure cooker and it's quite possible, even easy to over cook them. They start breaking apart a lot, though if you blend those you get a nice soup.

My favorite Python package manager is github.com/JuliaPy/Cond...

On larger matrices it does quite well on an apple laptop
```
julia> A = rand(1000,1000); B = rand(1000,1000);
julia> @btime $A * $B;
7.033 ms (3 allocations: 7.64 MiB)
julia> using BLISBLAS
julia> @btime $A * $B;
5.211 ms (3 allocations: 7.64 MiB)
```
Though the problem size is different

I think we are building/using it wrong, because their own results show very different results github.com/flame/blis/b...

Somehow Hanlon's razor (Never attribute to malice that which is adequately explained by stupidity), doesn't apply to him, he managed to rise above it an be stupid and evil

😅I’m not even an Academic. But yeah it could’ve been worded better, there are much better answers than mine in that thread

We ship OpenBLAS, but LBT allows us to change blases (blasi?) at runtime which is really convenient for benchmarking

While most people seem to be researching the behavior of LLMs I do enjoy that some people still have the time for real subjects like LMM

👋