I would guess “us vs them” mentalities are not good for productive societies

I think you're confusing the expected behavior of the product offerings. Every user gets their own VM for free. would you be similarly convinced an attack has happened if AWS gave you a remote shell to the instance you rented?

only sorts numbers? wouldn't radix be much better?

this is basically the only thing pre-training teams work on in labs. compute efficiency is the metric, the assumption that scaling = intelligence is considered a given.

they say they're looking at base models, so I think it's fairly compared as written.

this is a nice and concise writeup. what's striking to me is that these techniques really have not changed in /years/. sure, precision has become slightly lower, spec decoding acceptance has gotten slightly better and the complexity of parallelism is trickier with mixture of experts. but no new concepts in a very long time!

the absolute most impactful improvements for inference comes at architecture design time. I firmly believe everyone who cares about impacting model efficiency should look there

I think the biggest net new recent technique is P/D disaggregation. And that spec dec is very different now especially post DSpark/DFlash.

But overall yes the fundamentals of LLM performance optimization have been remarkably stable over the last few years.

perhaps its unfair to say this in hindsight, but it's a fairly straightforward application of little's law that's been around for some time

https://arxiv.org/html/2401.09670v2

It's also the hardest point at which to try to work, because when you change the model architecture you need to completely retrain from scratch.

this is Qwen3.8 max, right? https://qwen.ai/blog?id=qwen3.8

That's my bet, I doubt they'd have the resources to have trained anything substantially larger

they explicitly say this is qwen3.8 max in the hugging face page

are these all uniform quantization? or mixed and matched by layer (can't tell from the naming scheme)

this is cool but like, are we just vibe coding NAND burners at this point? these decode times don't really tell the whole story, because prefill becomes the bottleneck.

half an hour to process 10k tokens on an M5 seems... not great

This is how progress happens, someone gets to 3t/s, the next person gets to6/s and eventually we get to 100t/s.

People like this person are laying the foundations.

There is a limit how much you can squeeze out of given hardware. Betting on it being closer to 100t/s than 6t/s is only that, a bet.

On my 4090 setup I'm getting 86t/s on a 12B Q6_K quantized model running entirely in VRAM.

The current GPUs are optimized for processing huge numbers of triangles per second. There are three things at play here:

1. the organization of the data being sent to the GPU to optimize throughput;

2. the speed at which the GPU can read that data from its VRAM;

3. how many triangles it can process in parallel by using individual compute units.

I suspect that given parallel improvements for neural networks, we'll see similar improvements:

1. optimizing the structure of the weights in the model for efficient access by the CPU/GPU/NPU/TPU;

2. efficient access of data strides (matrix rows) in the memory, e.g. being able to read multiple 2x2 matrix values in one clock cycle, or stepwise pairs of values (a(i,j), b(j,k)) needed for matrix multiplication;

3. parallel compute for matrix and tensor multiplication and other operations needed by neural networks.

If you've got X GB of weights in slow access memory (be it RAM vs VRAM or SSD vs RAM) and Y GB of fast memory then no matter what, if you want to use them you'll need to transfer X-Y GB and will be bound by memory throughput. You can try to reduce number of activated weights, but how much can be gained that way is speculative so far.

Not great for coding, or realtime agent interactions. But for background processing tasks overnight? Seems like it’d work pretty well

I'm pretty sure one can rent a GPU for a few minutes with the electricity cost of leaving an M5 overnight.

Sure. One could. But then one wouldn’t be in control of every step of the process.

Domestic electricity is free nowadays, certainly for most of the year, as solar plus battery covers your usage for a tiny percentage of the cost of your house.

Only if you don't count the cost of the equipment and installation.

Or the cost of the house.

Given the cost of a building is far more than the cost of generating enough power for that building it doesn’t really matter

SSD NAND reads are nearly infinite. Still makes me uncomfortable, but writing is what kills. There's a reason SSDs are rated by TBW, not TBR.

agree, prefill is the weak spot right now. it goes through the same per-token path as decode, which is dumb for long prompts. The fix is on the list: during prefill we can batch the expert reads for the whole prompt per layer instead of per token, that amortizes the IO a lot. until that lands, honest answer is this is good for chat-length stuff, not for feeding it a 10k token document.

there's wifi ah, which runs on 900mhz band and has the same 30dbm limitation

I've used it with some raspberry pis to create hi-fidelity walkie talkies it's quite pleasant.