I've got a very real world use case I use DistilBERT for - learning how to label wordpress articles. It is one of those things where it's kind of valuable (tagging) but not enough to spend loads on compute for it.
The great thing is I have enough data (100k+) to fine-tune and run a meaningful classification report over. The data is very diverse, and while the labels aren't totally evenly distributed, I can deal with the imbalance with a few tricks.
Can't wait to swap it out for this and see the changes in the scores. Will report back
I'd encourage you to give setfit a try, along with aggressively deduplicating your training set, finding top ~2500 clusters per label, and using setfit to train multilabel classifier on that.
Either way- would love to know what worked for you! :)
You can solve this by training a model per taxonomy, then wrap the individual models into a wrapper model to output joint probabilities. The largest amount of labels I have in a taxonomy is 8.
Grey market, not black. It's been several months since I've talked to anyone in the space but full-chain reliable quiet Chrome exploit packages were high six figures, with discussions starting about bugs reaching 7 figures imminently, and the people I talked to might have been talking that down (or talking it up).
Again, remember that grey market payouts are tranched, so you could get 3x more than Google would pay, or you could get 0.5x, and for much more work.
Google security team is really good, however sometimes things are controversial because certain bugs gets ignored in MS-way which is famous for not paying/not fixing.
I felt the same when implementing OpenID connect flows according to spec. It uses the browser in creative ways ;) Especially the device flow, absolutely insane complexity for what it is.
They're just public/private keypairs that are generated either by a device (whether it's part of you phone, computer, or hardware key), browser, or password manager. I do agree that it can be a bit of a pain when it comes to multiple managers trying to offer to save/respond to a passkey, but otherwise it's a fairly straightforward exchange.
> They're just public/private keypairs that are generated either by a device (whether it's part of you phone, computer, or hardware key), browser, or password manager
Now imagine saying that sentence to a person outside tech
Why would you give the technical explanation to a person that doesn't want the technical explanation? To the person outside of tech, passkeys are just your phone has a really good password and fills it out for you. Just use that and don't bother having to remember (and forget) another password.
A monad is just a monoid in the category of endofunctors, what's the problem? Ape holders can use multiple slurp juices on a single ape, so if you have 1 astro ape and 3 slurp juices you can create 3 new apes.
The biggest speed-up in uv comes from the way it uses caching and hard links. When you install a package into a virtual environment uv uses a hard link to a previously cached version rather than copying files.
Using Rust is responsible for a lot of speed gains too, but I believe it's the hard linking trick (which could be implemented in any language) that's the biggest win.
There's also the assumptions around wheels having consistent metadata (which isn't required in older versions of the specs), so as long as you don't need to support the long tail of packages, then those assumptions are fine (it's worth noting uv weren't the first to make these assumptions, poetry did this too, but pip does not).
I am happy to pay more money given the companies goals, and that extra money is an investment to me. If I didnt buy it they have one less sale, and I won't have contributed to making the world have more companies like framework. I have hope others are doing the same despite them not being the cheapest.
If they stop delivering, ill not buy their next thing, and ill be sad.
> How are UK citizens voting for deteriorating services?
- Voting Tory, you get cutting of public health services (e.g NHS) and and Brexit.
- Voting Reform, you get the same plus a healthy dose of xenophobia and nationalism (it's thinly veiled-fascism, really).
- Voting Labour, you get the same cuts but with a nice label of "progress", so you can feel good about yourself. All the while flat prices are skyrocketing and your doctor's appointment might not ever happen, but there is always money for militarizing police and for sending abroad some bombs to kill Palestinians.
If people did want improved public services, they should have voted for somebody like Jeremy Corbyn, who had the decency to distance himself from the absolute disgrace that Labour has become, or at least for the Greens/Socialists. But people did not, in fact, want improved public services, not really.
They preferred to either vote for the xenophobe thugs of Reform, the xenophobe elitist Tories, or the pretend-liberal oligarch/capital-loving "Labour", none of who have any incentives to improve public services. You get what you vote for.
The great thing is I have enough data (100k+) to fine-tune and run a meaningful classification report over. The data is very diverse, and while the labels aren't totally evenly distributed, I can deal with the imbalance with a few tricks.
Can't wait to swap it out for this and see the changes in the scores. Will report back
reply