weebull

It's a prompt. Writing anything for that purpose is code gardening, but hey it's in rust so I'm supposed to be excited?

That, in essence is the problem "X in rust" normally means "I've written something of low value IN RUST. Gimmee upvotes". Come back when the project is interesting regardless of the language.

> Completely guessing, but the thought comes to mind that Rust was (still is? I don't keep up these days) portrayed as The C Killer. So to specify "written in Rust" was to imply "I wrote something low-level without C/C++" or "I rewrote Popular C Tool in Rust" in the name of memory safety.

As the OP, this is a big part of it. Rust might be a great language, but people announcing they rewrote grep (for example) in rust doesn't mean they've done anything special. The rust language team did something special. The person making the announcement just took somebody else's idea and reimplemented it for very little reason.

Great! You learnt enough to reinvent the wheel. Now do something useful.

Of course. It's obvious why....

The bath tubs are too small.

Rocm hasn't been the the problem in my experience. It's things like pytorch on rocm. It uses hippify to convert cuda code to hip rather than having native support.

It "works", but it's an obvious second class citizen, and the developers don't seem to take bug reports / PRs seriously.

Am I the only one who is getting tired of "It's X in rust" type projects? It's making me dislike the community.

Rust is not a user feature, it's an implementation detail.

<cue people telling me I should consider Rust a feature>

That's exactly my point. Cross dependencies is exactly the problem I want CI to solve.

GitHub actions would suck whatever you "configured" them in, because you are trying to describe a program in a data structure.

Ansible makes the same mistake, as do countless other tools.

You're safe because you're right.

Monorepos are a problem born of CI which can't cope with cross project dependencies properly. People have solved the problem by pumping everything together, but it's the wrong answer.

Fix CI and the problem goes away.

ASCAP Vs Open AI in 3...2...1...

So Ryzen? Same fabs and nodes as Apple.

Calling Apple Mx "a modified phone" CPU is doing it a disservice. It's a synergy of mobile design with desktop design.

Interesting, when did it move?

It's the point where you're no longer an individual contributor. Running a team and keeping people on target is difficult remotely. Consensus is difficult remotely. Feedback is difficult remotely.

Getting your head down and coding on your personal goal is easy remotely.

I think some ICs like that it's harder for people to say to them that they are driving in the wrong direction, even if it's still true.

If somebody can show me a coding task that LLMs have successfully done that isn't an interview question or a documentation snippet, I might start to value it.

Spending huge amount of resource to be a bit better at autocompleting code doesn't have value to me. I want it to solve significant problems, and it's looking like it can't do it and scaling it to be able to is totally impractical.

> In aggregate, training all 9 Code Llama models required 400K GPU hours of computation on hardware of type A100-80GB (TDP of 350-400W).

That is: * 45⅔ GPU years * 160 MWh or... * 45 average UK homes annual electric consumption * 18 average US homes * 64 average drivers annual milage in an EV.

...and that's just the GPUs. Add on all the rest of the system (s).

Ellama for Emacs look promising, but I only tried to install it this morning.

I'm honestly failing to see the utility for LLMs, because the context for any given problem is far too small, and we're already at 33B parameter models. They just don't seem to be a technology that scales to an interesting problem size.

There are these things called Loras (Low Ranked adaptions) which do exactly this.

"A language model with billions of parameters may be LoRA fine-tuned with only several millions of parameters." -- Wikipedia

Machine learning can tackle this for sure, but that's surely separate to LLMs. A language model deals with language, not logic.

Could also be Israel, Switzerland, Greece, Brazil, or any number of other places

The problem there is with PyTorch. They've got bugs in their AMD build flow that people keep pointing out to them, but because it's people with consumer cards which are "unsupported" (which just means AMD won't personally solve your problems with those cards, not that they aren't intended to work) they close the issues / PRs without fixing the bug.

AMDs wording isn't helping either of course.

It's Hawaii. They're literally sitting on an infinite energy supply and have almost continuous sun (apart from nights).

LLM? Probably ok. Image diffusion? Not so hot, but better than CPU.

Not OP, but I read their words as neither of the options you suggest.

Claiming "I was provoked" as justification means some arbitrary line was crossed. That line will be different for you and me, and as such it's not an objective measure. Ergo, it is disconnected from the objective reality of the situation, and much more connected to the subjective interpretation of the situation.

The "attempt to save critical data" is a pretty major difference.

In that situation, you just move it.

...which is why the 75% using VS code is a bad thing.

Slaves. The answer is Slaves.

It's because Americans use the term "gas" to mean the liquid fuel they put in their cars. In the UK it's just "gas".

It would be far better if we called it methane, but that's not really accurate either. It might contain ethane, propane, butane pentane and hexane. A lot of that stuff is condensed out before use, but not all.

"PyTorch uses the best backend it has available. If that's CUDA, it will use CUDA. It will also use ROCm on AMD GPUs and optimised BLAS libraries on CPUs. These backeds are selected at compile time though, so you need to install a version of PyTorch with optimised support for your hardware. "