This quantization has a small order of magnitude improvement on memory and compute requirements, how can it be slower?
And all that while retaining quality.
85 karma · joined April 30, 2014
This quantization has a small order of magnitude improvement on memory and compute requirements, how can it be slower?
And all that while retaining quality.
- use Ising machines to describe a certain problem into clauses, storing system state (e.g. spin of something) in variables
- then use a neural network layer where each neuron determines the value of one clause
- then for each state item, use the neuron output to determine if flipping that state would improve the overall system score
- and then use FN-like "noise" to determine whether to flip or no
If the energy landscape of the problem is pretty local, this is guaranteed to find a good solution to the system, using way less compute than brute-forcing it.
BTW I see that deepseek V4 pro is trounced by it's little flash brother? Any ideas as to why?
A CyberTruck will charge just fine from anything that can generate the proper AC or DC, no phoning home needed. Many home solar installations can work off grid and charge your car.
I image these displays could have color sensors attached to self-calibrate.
Or the variability is low and all you need is very precise voltages.
I think the first versions will be RGB displays with fixed colors, just no longer needing mass transfer. You could use tens of subpixels per pixel, reducing all worries about color resolution.
Make these into e.g. 1x1cm mini displays and mass transfer those into any desired display size.
I no longer needed my glasses to drive or go to the movies. Granted it was only -1 but still.
Now a decade on, my diet isn't so clean any more, far away details are somewhat blurry but still better than they used to be. The biggest impact is the amount of sleep I had the night before.
> We introduced BtrBlocks, an open columnar compression format for data lakes. By analyzing a collection of real-world datasets, we selected a pool of fast encoding schemes for this use case. Additionally, we introduced Pseudodecimal Encoding, a novel compression scheme for floating-point numbers. Using our sample-based compression scheme selection algorithm and our generic framework for cascading compression, we showed that, compared to existing data lake formats, BtrBlocks achieves a high compression factor, competitive compression speed and superior decompression performance. BtrBlocks is open source and available at https://github.com/maxi-k/btrblocks.
Although it is my hope that rfc 17 eventually makes it through: https://github.com/wmertens/rfcs/blob/master/rfcs/0017-inten...
(Of course, I would also like it to be fast, so that will probably not happen in the next 5 years)
The article doesn't say if their port is also open source. There's a blog post here but I don't see any patches. https://riscv.org/blog/2021/11/how-alibaba-is-porting-risc-v...
Actually, yes, that would indeed be a good use of the concept - anchor the generator to a submerged bouey (so as to not disturb shipping) and it will always be pointing in the right direction and generating. Nice!
Way less logic and parts needed vs Makani
If they would transition to this setup, there's nothing for regulators to point at for antitrust enforcement.