tehsauce

8 ms·

91.
by tehsauce 3y ago
Apple's new M2 Max has a neural engine which can do 15 trillion flops. Nvidias's A100 chip (released almost 3 years ago) can do 315 trillion flops. Apple is not going to close this 20x gap in a few years.
92.
by tehsauce 3y ago
“Differentiable programming” is nice term which covers the broad application of NN/deep learning techniques to special domains
93.
by tehsauce 3y ago
It’s better to compare against an ML framework than to maximum theoretical flops because sometimes it’s not possible to reach. These models are often limited by memory bandwidth rather than flop capability.
94.
by tehsauce 3y ago
MAE ViT-H is available freely, in fact the entirety of the SAM pipeline is open source, and they even released the dataset.
95.
by tehsauce 4y ago
In the last section on hash tables, I was wondering how on earth they make hash tables work well on the GPU. The answer: “What do we do about hash collisions? Nothing—the training process will automatically find an encoding that is robust t
96.
by tehsauce 4y ago
A great summary of this line of research! I think it should have the (2022) label.
97.
by tehsauce 4y ago
I haven't seen a single example from this model that demonstrates video with any time of time continuity. It appears every frame is independent to each other.
98.
by tehsauce 4y ago
I tried to search for CuLitho on youtube on got quite surprising results
99.
by tehsauce 4y ago
the machines will pay you much more to fight for them
100.
by tehsauce 4y ago
Wow I just looked up the second last one, what a crazy event. https://en.wikipedia.org/wiki/1965_Philippine_Sea_A-4_incide... "The attack jet fell over the side during a training exercise while being rolled from t
101.
by tehsauce 4y ago
This is one of the best analyses of gpt4 Ive read so far. Besides potentially including the visual aspect, I wonder if part of the reason it has trouble with harder problems is that it’s been tuned/prompted in a suboptimal way. The ad
102.
by tehsauce 4y ago
The author of this post seems to have put their own watermark these images which they did not create or cite? For example I recognize some made by inigo quilez.
103.
by tehsauce 4y ago
Did an admin just edit the title from: "YC is asking for a bailout" to: "Urgent: Sign the petition now" ??
104.
by tehsauce 4y ago
Shamelessly sharing some research I worked on a few years ago- Using ML to learn cellular automata which simulate a given system (uses neural cellular automata). Code: https://github.com/PWhiddy/Growing-Neural-Cellula
105.
by tehsauce 4y ago
This just blew my mind, never realized that!
106.
by tehsauce 4y ago
Here's a video summary of the research: https://youtu.be/U93bUQ1roiw
107.
Ask HN: When will Nvidia H100 GPUs be available?
1 points by tehsauce 4y ago|0 comments
108.
by tehsauce 4y ago
How exactly is this different than just importing the openai library in a regular backend template?
109.
by tehsauce 4y ago
How did they get to use the joe rogan voice though? It seems that one isn’t public?
110.
by tehsauce 4y ago
stable diffusion
111.
by tehsauce 4y ago
Yeah! For demo and testing purposes. Feel free to play around/build toy apps with it. There is a small global rate limit for all unauthenticated users though so don't rely on it :)
112.
by tehsauce 4y ago
Shortly after stable diffusion was released, I realized that an enormous number of non-ML people were suddenly interested in using an ML model. However, APIs are insanely expensive and not very developer friendly, and running it yourself re
113.
by tehsauce 4y ago
I also didn't like the abstract. It feels more like what the introduction section of the paper should be. Instead here they should summarize what their key finds or results were.
114.
by tehsauce 4y ago
I’m not sure what you mean saying “gimp being behind the times” compared to photopea. Gimp absolutely blows it out of the water in terms of being a useful tool.
115.
by tehsauce 4y ago
Prepending “Hi I’m ChatGPT” to the prompt of another model is hardly a recreation :/
116.
by tehsauce 4y ago
This looks great! Thanks for sharing.
117.
by tehsauce 4y ago
I would like to support dreambooth, if there is a way to store the fine-tuned models more efficiently. The challenge is that each trained model is quite large and a bunch of models can't fit into one gpus memory at once.
118.
by tehsauce 4y ago
The GPUs are rented from vast.ai The individual machines aren't as reliable or well-integrated with other cloud services as typical cloud machines, but multiple can be put behind a queue to create a highly reliable service.
119.
by tehsauce 4y ago
Hm, what device/os are you browsing on? The site should be mobile-friendly except for the account page. Also please feel free to reach out by email or discord.
120.
by tehsauce 4y ago
Very cool! I just released a tutorial on how to do this using the computerender api! https://github.com/computerender/tutorials/tree/main/python/... If you’re interested in saving money on expensive

More ›