pizza

32 ms·

511.
by pizza 3y ago
Here's a simple calculator for LLM inference requirements: https://rahulschand.github.io/gpu_poor/
512.
by pizza 3y ago
For my use-case, I wanted whole-paragraph embeddings (iirc these are trained on max-256-token sentence pairs) Their simple suggestion to extend to longer texts is to pool/avg sentence embeddings - but I'm not so sure that I want t
513.
by pizza 3y ago
One word of advice, the smaller your library, the fewer fires you'll have to put out in the future (bugs) and the lesser the energy spent on future rewrites
514.
by pizza 3y ago
nvidia specific, best write up I know: https://timdettmers.com/2023/01/30/which-gpu-for-deep-learni... Across vendors, generally, Nvidia still dominates currently. People are adding more support into ML libra
515.
pystrace: Library and CLI tool for collecting strace events(github.com)
1 points by pizza 3y ago|0 comments
516.
by pizza 3y ago
Using this, is it possible for live running application state on one device to be sent to another device, esp if running a different OS?
517.
by pizza 3y ago
Do you have an outlet for your expectations? Is that outlet also the people who are adjacent to your burnout? Are you sure your model of what is asked of you is not excessive compared to what they were asking? Is there a way you can negotia
518.
by pizza 3y ago
Kant would like a word with you about your point on whether people themselves understand the world and not just the format of their perceptions... :) I think if you're going to be strict about this, you have to defend against the point
519.
Orange Skies Day(en.wikipedia.org)
1 points by pizza 3y ago|0 comments
520.
America's First Commercial Carbon-Sucking Facility Opens in California(news.slashdot.org)
3 points by pizza 3y ago|0 comments
521.
Sleeper Hit(en.wikipedia.org)
3 points by pizza 3y ago|0 comments
522.
What is the origin of phrase "for fun and profit"?(english.stackexchange.com)
2 points by pizza 3y ago|0 comments
523.
by pizza 3y ago
FWIW the paper title is focuses on quite a different conclusion than the submission title: Pretraining Data Mixtures Enable Narrow Model Selection Capabilities in Transformer Models
524.
by pizza 3y ago
How about 'post-processing that increases the information of an output signal beyond the information of the input signal', i.e. something that violates the data processing inequality.
525.
by pizza 3y ago
I think you would prefer something like a support vector machine over k-nn for better learned point-wise local rescaling (you could also use eg ISOMAP or MDS to try to linearize the manifold), which you could also use w various tda methods
526.
by pizza 3y ago
Would love a bluetooth muxer via rpi. Have a few devices that keep conflicting with one another and acquiring the connection to my bluetooth headset. I have to go through the hassle of navigating to the offending device's bluetooth set
527.
by pizza 3y ago
I thought they were asking about whether a coordinate system with basis elements e_1, e_2, ..., could be re-parameterized after rotation, and whether the re-parameterizations are infinite. The answer is simple via geometric algebra: yes. Gi
528.
by pizza 3y ago
Check out the article "A Unified Mathematical Language for Physics and Engineering in the 21st Century" [0]. [0] http://geometry.mrao.cam.ac.uk/wp-content/uploads/2015/02/00...
529.
Command Line Interface Guidelines(clig.dev)
2 points by pizza 3y ago|1 comments
530.
by pizza 3y ago
Another link https://webbtelescope.org/contents/news-releases/2023/news-2...
531.
by pizza 3y ago
ty
532.
by pizza 3y ago
My mistake, thanks
533.
by pizza 3y ago
Somehow this reminds me of pre-y2k Nick Land ( Meltdown [0] etc.) (Hope that doesn't correlate with the emergence of a 'post-y2k-Landian' Andreesen-Horowitz!) [0] https://www.youtube.com/watch?v=fiaWsgtJrNI
534.
by pizza 3y ago
Nice post. Also somewhat surprised Core ML doesn't have an implementation of GeLU since it's so widely used at this point (and not particularly challenging to re-implement, were I the speculative type..)
535.
by pizza 3y ago
edit: I'm not sure what jdkwkbdbs means (dead-banned) by "LLMs don't. ML works pretty well." (well, I do); LLMs solve certain tasks -- and really interesting ones, at that, too -- at certain average cross-entropy loss le
536.
by pizza 3y ago
Why do you say it doesn't work well? I thought it did?
537.
Nvidia 2024-2025 data-center roadmap lists GB200 and GX200 next-gen GPUs(videocardz.com)
2 points by pizza 3y ago|0 comments
538.
Nvidia's RTX 5000 Ada Now Available: AD102 with 32GB of GDDR6(tomshardware.com)
51 points by pizza 3y ago|35 comments
539.
by pizza 3y ago
These models are pretty much always extrapolating [0] Whether the extrapolation is crude/low-rank or astute/high-rank is a question of memorization vs generalization. That gets into the question of whether or not the model is over
540.
History Data Structures(gist.github.com)
3 points by pizza 3y ago|0 comments

More ›