jerpint

16 ms·

241.
by jerpint 3y ago
The gist is only to create the dataset not to fine tune
242.
by jerpint 3y ago
The ugly truth is that the competition is largely worse. I recently wanted to watch a show only available via a third party thru Amazon prime, so I purchased a subscription (on top of prime), only to be served a show with adds every 10 minu
243.
by jerpint 3y ago
It’s pretty much the driving force behind all open source machine learning demos these days, it was acquired by hugging face a few years ago and they do an incredible job maintaining it
244.
by jerpint 3y ago
The author didn’t mention the arguably most popular python frontend library: gradio
245.
by jerpint 3y ago
Google is also less and less “useful” because social media has essentially turned most of online content into content that cannot be indexed by google
246.
by jerpint 3y ago
It’s similar to preferring guitar hero because a real guitar has too many frets and too many chords to learn
247.
Growing Snowflakes with Neural Cellular Automata(jerpint.io)
2 points by jerpint 3y ago|0 comments
248.
by jerpint 3y ago
I’m getting married tomorrow, and doing tables is no easy feat. The list constantly evolves, people come/cancel last minute, and sometimes you can play on multiple relationships for seating (e.g. one couple might be family but also fri
249.
by jerpint 3y ago
It is an awful idea because it will make most people sad not to be next to the ones they love, like spouses and friends
250.
by jerpint 3y ago
Yes but they are not trained to explicitly encourage similar texts to be semantically similar, only to do next token prediction. In embedding models a contrastive loss is used to minimize distance between pairs of semantically similar conte
251.
by jerpint 3y ago
It would be interesting if that was at all a confounding variable , one thing for sure is its impact has been under reported
252.
by jerpint 3y ago
The internet has allowed us to interact in ways that were inconceivable at the time; think communication and speed of information for one. When agents start being more reliable I think we will start seeing applications we couldn’t possibly
253.
by jerpint 3y ago
Yesterday I wanted to watch a show, which was available on prime. When trying to view it, I had to pay an extra 5.99$ for another subscription to watch, which I paid. I then got served adds every 10 minutes during my show. Streaming service
254.
by jerpint 3y ago
Don’t forget a bunch of useful git aliases alias gcm=git commit -m alias gaugcm = git add -u && gcm
255.
by jerpint 3y ago
I use a variation of this to enable my conda environments - I always name the conda env the same as the top level git folder and can then quickly enable my env with an alias
256.
by jerpint 3y ago
If you’re looking for an open source RAG solution, try our library: https://www.github.com/jerpint/Buster
257.
by jerpint 3y ago
I asked for South Park and it complained about copyright so I asked for east park where any connection to South Park would be completely accidental and happily obliged to create ~south~ east park
258.
by jerpint 3y ago
Not available in Canada
259.
by jerpint 3y ago
Yannick kilcher YouTube Channel is filled with very relevant paper reviews that are very helpful
260.
by jerpint 3y ago
You have to ask yourself not “how impossible” it is , rather how much effort was put into mitigating this. The answer is probably a nominal amount but obviously not enough
261.
by jerpint 3y ago
Tokeninazation make this kind of “dyslexia” very unlikely
262.
by jerpint 3y ago
That’s very beautiful, thanks for sharing and happy to read you have no regrets :)
263.
by jerpint 3y ago
RAG and LoRA are 2 separate things - I’m RAG, you don’t do any fine tuning , rather you use in context learning
264.
by jerpint 3y ago
LoRA stands for low rank approximation - it’s a very clever way to fine tune large models with a fraction of the compute, the idea being that you need much less capacity to estimate the finetuning
265.
by jerpint 3y ago
Just ran this on my phone through a browser, this is very impressive
266.
by jerpint 3y ago
I found the silence on HN around these events deafening when the Oct 7th attacks first happened (before Israeli retaliation). What is depicted here is tragic. There were many other tragic depictions then too. It seems fair to discuss both t
267.
by jerpint 3y ago
what do we do about javascript?
268.
by jerpint 3y ago
You could set up something similar using gradio quite easily: https://www.jerpint.io/blog/model-inference/
269.
by jerpint 3y ago
Solved it all last year, looking forward to finishing this year’a puzzles!
270.
by jerpint 3y ago
The neural network module looks identical to PyTorch

More ›