pama

32 ms·

481.
by pama 2y ago
I’m afraid that the pipe character as a separator of columns will lead to inevitable errors in scripts or copy/paste with more parts of the output having a chance of being executed than with regular ls.
482.
by pama 2y ago
The repo mentiones a Karpathy tweet from Jan 2023. Andrej has recently created llm.c and the same model trained about 32x faster on the same NVidia hardware mentioned in the tweet. I dont think the perfomance estimate that the repo used (b
483.
Mexico man dies from first human case of bird flu strain H5N2(theguardian.com)
38 points by pama 2y ago|6 comments
484.
by pama 2y ago
Nothing original to see here and nothing reformulated to be of practical importance other than a clickbait attempt. If this is your first trap of reading absurdist/populist LLM literature consider yourselves lucky. If you care about th
485.
by pama 2y ago
Because nobody else mentioned it, elisp is here to stay and a very useful tool to have around if you use Emacs. Not the one-true-Lisp for sure, but a practical way to make your everyday life more enjoyable and efficient and enough to satisf
486.
by pama 2y ago
Yes reinforcement learning. Apologies for the jargon.
487.
by pama 2y ago
This could simplify RL on ncurses codes.
488.
by pama 2y ago
Without the accelerator addon the Pi5 could do smaller models at passable speed, so I wonder if an accelerator can make this faster and allow larger models. https://robleshermoso.com/posts/ejecutando-llm-raspberry-pi-..
489.
by pama 2y ago
If it helps think of the first author as the lead engineer or the CEO, and the last author as the board or the VC. In some areas of science (or some teams) the last author is closer to a CEO, in others closer to a VC (they almost always ha
490.
by pama 2y ago
Can one run fast LLM inference on these?
491.
by pama 2y ago
Mamba is technically a recurrent neural network (and typically decodes as such), simply of a constrained architecture that among other things keeps the norms of its matrixes finite independently of gating. I think that the above paper conf
492.
by pama 2y ago
The last figure in this paper is a huge disappointment when you consider how reality meets such theoretical arguments. A typical mamba model would have 100 layers (maybe 60 for snaller ones). That figure scales up to 4 layers and these are
493.
by pama 2y ago
Has anyone tried training yet and are there any obvious pitfalls for multi GPU training like there were in mamba-1?
494.
by pama 2y ago
What was the final loss? Is this hardware available for rent somewhere?
495.
by pama 2y ago
Did you reproduce the evaluation as well?
496.
by pama 2y ago
I authorize you to be authentic!
497.
by pama 2y ago
The high level strategy would be to get the browser tab contents and then ask a local LLM to organize the tabs based on their full content. If you run selenium or some developer versions of certain browsers you might be able to source the f
498.
by pama 2y ago
Unfortunately this statement would not offer sufficient legal protection, so the original authors would have to be convinced to give up their previous rights and change the upstream copyright (and huggingface should update their repo licens
499.
by pama 2y ago
The article does discuss these.
500.
by pama 2y ago
Fully agreed.
501.
by pama 2y ago
I thought I wasnt alone but perhaps I live in a sparsely populated alternative history where Emacs gets simpler over time. Once you know the basics they dont change. Some more advanced tools gradually simplify or improve but it takes years.
502.
by pama 2y ago
Sorry to hear. Codes have bugs that evolve over time. I hope you try magit again some day. It is sometimes hard to remember to apply correct quoting in a shell, so if in Emacs you can also use dired for dealing with such errors in less risk
503.
by pama 2y ago
It is not yet a glofiried ssh machine, so that is an 80% no when it comes to work hardware for me. But give me a docking station with my homemade keyboard, fancy monitor, mike, and headphones, and let me use the GNU toolchain locally to get
504.
by pama 2y ago
A reasonable guess is that some entities are storing / collecting attacks patiently waiting for one big event. Having smaller constant incidents only helps strengthen the opponent over time, thus making it harder to deploy a coordinate
505.
by pama 2y ago
> Other than safety and the like... This statement does a lot of work. Safety without compromise in speed on modern architectures is rather valuable. I dont think you understand the importance of that statement and why it is being used i
506.
by pama 2y ago
Awesome. Do you happen to have a benchmark against the latest (v9.1) cuDNN implementation?
507.
by pama 2y ago
I’m happily impressed that non experts find this area intriguing. I hope more people start to work at the interface of biology, chemistry, and AI.
508.
by pama 2y ago
It is absolutely amazing. Thank you to everyone at OpenAI!
509.
by pama 2y ago
Thanks! Typically I don’t like solutions that involve extra environment variables as things inevitably get messy over time, but your use cases are clean, simple, and common enough that it’s nice to have them documented in one place in this
510.
by pama 2y ago
It is possible that this second recursion doesn’t hold at N=3 (because there is enough of the density covered by the fraction of real polynomials), but then only starts at higher degrees (it will eventually be easier because the N-2 degree

More ›