pama

53 ms·

751.
by pama 3y ago
Here is a breakdown of some of your questions: 1. Atomicity of Rename Operations: On many file systems, the process of renaming a file is atomic. This means that the operation either completes in full or doesn’t take effect at all. This mak
752.
by pama 3y ago
There are good reasons for the Emacs default that relate to atomic operations in typical file systems, reliability in shady hardware, or power losses and recovery. The argument of the author about hard links breaking is a matter of choice;
753.
by pama 3y ago
Yes. The move rename operation is atomic in any reasonable file system. With the default behavior Emacs can always keep the previously known good data in place even under power failure, disk failures, etc. if the new file is corrupt you go
754.
by pama 3y ago
I agree. Emacs defaults are battle hardened and work well in exotic and regular hardware for all kinds of use cases. I modify defaults to remove the extra fluff (tool bars, menus, scroll bars, yes-or-no), but I see their value to new users.
755.
by pama 3y ago
I like backups as a reminder that some file is not under version control.
756.
by pama 3y ago
I think that the default backup behavior is useful: it fits the idea that the backup is more important than the new file, and it gives copy-on-write powers to hard links. If Emacs sees the file is version controlled it will not back it up
757.
by pama 3y ago
LED on oled stands for light emitting diodes, so they are a backlight.
758.
by pama 3y ago
One problem with pivot and its reverse operation is the poor terminology surrounding them. In addition to pivot itself, people used unstack, stack, gather, melt, reshape, transpose, … A single reversible concept like pivot_wider and pivot_
759.
by pama 3y ago
These plots show a persistent difference at 30 to 40 years past graduation. Society was very different when the first teams in their study, from 1970, graduated. Even though the classification of athlete vs non athlete is of course silly a
760.
by pama 3y ago
In the context of this article, the small/tiny models were 1--30 million parameters and the large model was 1.5 billion parameters. Efficient training of a 7-billion parameter model already requires algorithms for training across multi
761.
by pama 3y ago
Empirically, perhaps due to the lottery ticket hypothesis, one needs to wait a lot longer for small models to reach a useful loss, so if one is impatient it almost always pays off to train the largest model that a given compute configuratio
762.
by pama 3y ago
This is an up to date discussion of useful techniques for training large models.
763.
Fine-tuning Llama 2 70B using PyTorch FSDP(huggingface.co)
1 points by pama 3y ago|0 comments
764.
Everything about Distributed Training and Efficient Finetuning(sumanthrh.com)
3 points by pama 3y ago|1 comments
765.
by pama 3y ago
I don’t know for sure why the early pytorch team picked it, but my guess is due to simplicity and performance. NVidia optimizes CUDA better that OpenCL and provides tons of useful performance tuning tools. It is hard to match the CUDA perfo
766.
by pama 3y ago
The pics of the offer and pricing suggests uniform DNA info rather than names or passwords for some and raw DNA profiles for others. Also this from the article: “ The compromised accounts had opted into the platform's 'DNA Relati
767.
by pama 3y ago
Thanks. You are actually right that this new supercomputer might move the needle once it is in production mode. I will wait and see how it goes.
768.
by pama 3y ago
I don’t think you are reading this correctly. People could access (most importantly) the full raw DNA profile. And many of those breached were from people who opted in a “Relatives” feature even if their account was secure.
769.
by pama 3y ago
There is only limited empirical evidence of AMD closing the gap that NVidia has created in the science or ML software. Even when considering pytorch only, the engineering effort to maintain specialized ROCm along with CUDA solutions is not
770.
by pama 3y ago
Up to a point, TSMC could replicate or increase the throughput of its fabs if demand increased systematically, which is plausible as ASIC designs become more common. Solving the current problems of a company like OpenAI requires only a very
771.
by pama 3y ago
Yes. For a model within the limits of the head requirements, however, you wouldn’t be able to see a quality difference from regular attention. Non determinism is a performance price; regular transformers may also suffer from it depending o
772.
by pama 3y ago
Poppler and its integration with Emacs via pdf-tools has dramatically improved my PDF reading and annotation powers. Since I started using it about 5 years ago, for the first time in over 30 years I prefer reading and annotating papers in d
773.
by pama 3y ago
Of course it is their right, however I don’t understand their argument in this text. They can start Emacs in daemon mode and issue emacsclient -c to have their main graphical session. I not sure I see the added complexity.. if they need mu
774.
by pama 3y ago
The result is identical to regular attention in transformers but training can be about four times faster, so there is almost no reason to not use it.
775.
An introduction to the problem of the AI conciousness(thegradient.pub)
2 points by pama 3y ago|0 comments
776.
Sticking together makes bacteria nearly invincible(economist.com)
1 points by pama 3y ago|0 comments
777.
by pama 3y ago
Not sure if it helps in your case: for a package worth keeping, I find it useful to build a script out of the commands I used on the machine I first installed it, and then try the script on another machine and improve it as I go along.
778.
by pama 3y ago
I don’t know of any established correlation of dehydration to any of this. Our bodies evolved over periods of time when bouts of dehydration were much more common. I just thought that maybe I missed some recent counterintuitive discovery.
779.
Greece, battered a decade ago, is booming(nytimes.com)
2 points by pama 3y ago|1 comments
780.
by pama 3y ago
Why?

More ›