omneity

20 ms·

331.
by omneity 3y ago
Brut is raw. Brute or brutal is related to violence.
332.
by omneity 3y ago
This happens to be the plot of a famous french comedy show in the 2000’s. Not sure if I should spoil as it’s literally the very last revelation in the series.
333.
by omneity 3y ago
Found it. The claim is not as strong as I remembered though: https://twitter.com/yanaiela/status/1764816123898565020
334.
by omneity 3y ago
I can’t find the tweet handy but a researcher showed this behavior was actually present in several datasets such as The Pile. So it’s less likely to be emergent behavior and just the model doing its thing.
335.
by omneity 3y ago
What about Keras?
336.
by omneity 3y ago
I wish this was ported to macos as well to enable eGPU setups. The last compatible driver predates the first RTX series.
337.
by omneity 3y ago
You're thinking of latency vs bandwidth/throughput. You might not improve on the latency part (speed of light), but you can increase the bandwidth (amount of data transferred per unit of time) just like a highway with more lanes c
338.
by omneity 3y ago
How would something like this impact our understanding of the heat death of the universe? I always thought interstellar space is an inert medium. But it seems something “interesting” _can_ still happen.
339.
by omneity 3y ago
Not quite. "in" here as a prefix is not a negation thing but to _do_ something like "en" in "enhance" or "encapsulate". The word's actual latin root is "inflammare" which means to put s
340.
by omneity 3y ago
That checks out in principle, but given that P40 doesn't support NVLink, I wouldn't count too much on using six of them together in a performant manner. But yeah the best option remains an MI300 if you can afford that.
341.
by omneity 3y ago
I agree with you, that’s how it should be, but that’s not how it currently is. Looking at what’s available right now. You need 3 A100 40GB to get this amount of VRAM which will cost you way north of 20000$. Doing it with A6000s is still abo
342.
by omneity 3y ago
A friend just got an M3 Max with 128GB (V)RAM and he's extremely happy with it (AI workloads wise). That could be an option if you can run your simulations on macOS.
343.
by omneity 3y ago
Would this explain why I started seeing so many ads from HBM manufacturers with no product an individual like me can purchase? (I'm not even close to hardware, professionally) Not naming the company but seems like HBM manufacturers mig
344.
by omneity 3y ago
Not the GP, but from my personal experience and exposure to Kerberos and SAML, most customers have a very custom SSO setup, permissioning can be hell (especially if the company did it a long time ago, and they didn't onboard a new SSO-
345.
by omneity 3y ago
I found the language in this article to be very strong and it seems to me this discovery only rules out a specific glycine exchange mechanism to power life reactions. The paper contrastingly is rather descriptive and much less assertive, an
346.
by omneity 3y ago
This is super cool! I’ve been looking for ways to integrate Observable with my blog posts and make them more interactive and engaging. This might just be it. Thank you and congrats on the release!
347.
by omneity 3y ago
This is amazing. Until now I used https://colorbox.io which first originated at Lyft before becoming an independent project.
348.
by omneity 3y ago
Using percents rather than absolute figures would make the comparison much easier. Great work though, this is basically skill gaps and “arbitrage” opportunities!
349.
by omneity 3y ago
My personal website and blog[0] makes use of the Liechtenstein .li TLD. That makes it special (for me). We also use .co[1] at my startup since someone is squatting our .com 0: https://omarkama.li 1: https://monitoro.c
350.
by omneity 3y ago
You can see gloves in Steamboat Willie’s outro too, on both Mickey and Minnie.
351.
by omneity 3y ago
I "interviewed" that engineer recently, especially about GPT-4 and how it compares to LaMDa: https://twitter.com/OmarKamali/status/1726431417163206900 (take the answers with a grain of salt, some people
352.
by omneity 3y ago
Can you recommend how to meet them? Ideally with a certain level of skill, and who doesn’t try to fleece you.
353.
by omneity 3y ago
I was wrong indeed, this is technically called zero-shot prompting. Zero-shot prompting here means without an example, i.e. "show me a panda on a skate". The opposite would be few-shot prompting, "write me a limerick like the
354.
by omneity 3y ago
That would mean most likely that the video is the result of a prompt without any example. Illustrated examples from how Google does it here: https://blog.research.google/2023/11/zero-shot-adaptive-prom... EDIT: co
355.
by omneity 3y ago
My understanding is that's partly how it got started (basically in a BDSM-influenced artistic context)
356.
by omneity 3y ago
I played this with a friend at the Computerspielemuseum in Berlin. I felt really bad, they were so bad I kept winning, and their hand’s skin cracked bloodily and got bruised after a session. But they insisted to finish the game.
357.
by omneity 3y ago
HF has been pretty good for training in my experience. Both accelerate, deepspeed and other improvements are available and the ecosystem is vibrant. One thing you might want to look into is to write your own training loop rather than using
358.
Mixtral 8x7B on 24GB VRAM (2-bit quantized)(twitter.com)
5 points by omneity 3y ago|0 comments
359.
by omneity 3y ago
To be honest, HF compat comes at the cost of abysmal performance compared to an inference solution such as vLLM.
360.
by omneity 3y ago
This feels like a badly implemented homegrown ECS pattern. ECS is kind of like lisp. Every sufficiently complex game will implement its own bug ridden ECS framework (unless it uses one already)

More ›