sorenjan

20 ms·

361.
by sorenjan 3y ago
How much VRAM does it need? They mention that the largest model uses 1.4 billion parameters more than SDXL, which in turn need a lot of VRAM.
362.
by sorenjan 3y ago
Maybe someone that's trying to use a GPU to solve a particular problem doesn't necessarily also have the time, energy, knowledge, or interest to also first 1) find out how to construct the toolchain, 2) build said toolchain, and 3
363.
by sorenjan 3y ago
ncnn uses Vulkan for GPU acceleration, I've seen it used in a few projects to get AMD hardware support. https://github.com/Tencent/ncnn
364.
by sorenjan 3y ago
Funnily enough it doesn't work on their RDNA ("Radeon DNA") hardware (with some exceptions I think), but it's aimed at their CDNA (Compute DNA). If they would come up with a new name today it probably wouldn't inclu
365.
by sorenjan 3y ago
But OsmAnd already uses their own vector tiles.
366.
by sorenjan 3y ago
I remember seeing his M&M sorting machine several years ago: https://www.youtube.com/watch?v=ceGlMV4sHnk
367.
by sorenjan 3y ago
This reminds me of one of the examples of what you can do with Huginn: "Create Amazon Mechanical Turk workflows as the inputs, or outputs, of agents (the Amazon Turk Agent is called the "HumanTaskAgent"). For example: "O
368.
by sorenjan 3y ago
I recently saw an interview with one of the people doing those videos. He makes over $15k per month, so I doubt it will go away without Google interference, but they obviously take a sizable cut of the total earnings from the same videos as
369.
by sorenjan 3y ago
The US sought a quick peace in Ukraine, they even asked Russia to not invade even before Russia started their invasion. You can't get much quicker than before the events have taken place. It didn't work because Russia isn't i
370.
by sorenjan 3y ago
There are adapters available for USB C cables if you don't want to modify your computer but want to use the same charger. https://www.aliexpress.com/item/1005005233750368.html https://www.aliexpress.com
371.
by sorenjan 3y ago
I think chaiNNer might be a better comparison, although both are used locally to process images while this looks like it's meant to easily build an API. I think it looks neat, I think a lot of people will find this very useful.
372.
by sorenjan 3y ago
That sounds perfectly reasonable. I'm more worried about noise and heat than the cost though, but I guess that's not too bad either then. What's the latency like? When I've used generative image models the programs unloa
373.
by sorenjan 3y ago
What's the power consumption and fan noise like when doing that? I assume you're running the model doing inference in the background for the whole coding session, i.e. hours at a time?
374.
by sorenjan 3y ago
That's exactly what the Q-learning lab in this course does: https://www.ida.liu.se/~TDDC17/info/labs/rl.en.shtml
375.
Reference SATA Cable – A flagship SATA cable for PC audio systems(jcat.eu)
5 points by sorenjan 3y ago|3 comments
376.
The scam YouTube ads are getting better(old.reddit.com)
73 points by sorenjan 3y ago|85 comments
377.
by sorenjan 3y ago
This is very impressive, but their approach to generate the whole temporal duration at once limits it to short clips. I guess one of the next steps is to make overlapping "clips" that then becomes longer videos.
378.
by sorenjan 3y ago
Thank you, I'm not used to reading this kind of research papers but I think I got the gist of it now. Can this architecture be used to distill models that need fewer timesteps like LCMs or SDXL turbo?
379.
by sorenjan 3y ago
I was referring to the input image in the diagram, what is that and how is the output image generated from it? Is it 256x256 noise that gets denoised into an image? I guess what I'm really asking is what guides the process into the fin
380.
by sorenjan 3y ago
This is probably a stupid question, but what kind of image generation does this do? The architecture overview shows "input image", and I don't see anything about text to image. Is it super resolution? Does class-conditional m
381.
by sorenjan 3y ago
Maybe you should update the text on your front page: > Why not use Android file transfer? > Android File Transfer is Google's tool for transferring files from your Android to your desktop. > The tool has not been updated in a
382.
by sorenjan 3y ago
The Nintendo 64 was very advanced for its time.
383.
by sorenjan 3y ago
Looking at the top 10 list it's dominated by American tech billionaires. Weirdly Steve Ballmer is #10, even though he was "only" CEO of Microsoft, not the founder. So much for the argument that the wealth comes from innovatio
384.
by sorenjan 3y ago
Is there a big difference in the result from fine tuning a model compared to using a LoRA? I thought the idea with LoRAs was that updating all model weights are unnecessary?
385.
by sorenjan 3y ago
Google has held a couple of GPS precision challenges for smartphones, and Galileo recently launched their High Accuracy Service (HAS) for high accuracy Precise Point Positioning. Android has built in support for full GNSS phase measurements
386.
ReLoRA: High-Rank Training Through Low-Rank Updates(arxiv.org)
3 points by sorenjan 3y ago|0 comments
387.
by sorenjan 3y ago
Someone on Reddit got a really nice love story between a Chevy Tahoe and Chevy Chase from it. https://imgur.com/vfHGHW6 https://imgur.com/JSjNC2c https://old.reddit.com/r/OpenAI/co
388.
by sorenjan 3y ago
Why would he have any pull to make Youtube do anything? What's he gonna do if Youtube refuses, stop using Youtube? He needs them much more than they need him.
389.
by sorenjan 3y ago
If you're willing to make the textured 3D model yourself there are services that can 3D print in color. https://www.shapeways.com/materials
390.
by sorenjan 3y ago
You don't have to generate a mesh token by token like you generate text. You can map the latent space to parametrized 3D models, or generate SDF, or a NeRF, or something else. See for instance: https://blog.metaphysic.ai&#x2

More ›