Also FYI doesn't run inference on Apple GPU (only for training)

Hey there, yep we found that on Apple devices specifically running on CPU is fast enough that Metal support is not needed. Thanks for flagging this though, and if usecases that would benefit from Metal support come up we will be adding it to the binaries.

I tried this today for labelling - and for that task it was very bad MNLI was better - so you are going to need to match the use case for this pretty exactly. (at 29MB params one would expect that!) I'm obviously not saying labelling is a good use case :-) just adding a data point.

Jev has put the cat amongst the pigeons so suddenly everyone is looking at classifiers and encoder only models again.

My ideal model would be a general purpose LLM API that can answer classification questions and as it does so distils to an encoder only model so that the more classifications I do the cheaper it gets (i.e. the more it offloads to the classifier). If anyone ever wants to do this as a service do let me know, because it's just another piece of code to manage in each new project that needs classification.

Also a model that could do this internally would be nice :-)

Hey! Yeah I think for labelling the model would need to have much better world knowledge than its current size allows. Jev really is a very good model, I think it has a very strong place in the upcoming tech stacks. Really good suggestion to make a continuously distilled model, we are going to have to look into that one :)

Good luck with this model/product, in the excitement of LLMs people seem to forget applicability. I very much like to see innovation in this space, so well done!

Please have a 'readable version' option so I don't have to exhaust myself parsing the sites layout. I get that it's unique but most of us just want to work out what you're offering in 5-10 seconds of our time.

I strongly second this, although I must admit it loaded surprisingly fast for me as I'm on a mobile hotspot in the back of a car.

HN: This site looks like all the other slop, awful to read.

Also HN: This site is doesn't look like other sites, awful to read.

I'd bet that no one, not even the site's [human] creators, has ever read that homepage end to end. At best, it might have been handed over to a swarm of reviewer agents.

OMG blast from the past!

Correct me if I'm wrong but Jev itself works pretty much the same as encoder only models.

I think so, yes.

However, it might have fewer restrictions than a BERT and/or is smarter (whatever that means).

Well that benchmark is now saturated, what next. How fast you can hack the pentagon?

Why on earth would we want such a lock-in at this stage when there is no clear winner. This is an area in which I would encourage everyone to build their own (using OSS) on top of existing cloud infrastructure.

I tried and it's difficult to create a good agent harness.

There are no shortage of good harnesses out there, and you can re-use coding ones like pi, opencode, dsh even claude and codex. For most use-cases that is already the loop you need. If you're looking for something less like a coding agent then Vercel's Eve is okay too.

But you can't just plug those into your already existing stack easily, unlike a purely API-based solution no?

To have a really good harness, it seems that you need a separate one for every different project you are working with.

Congratulations!!! You win what’s left of the internet - just ask Claude for your prize! Motrin I’ve had this week.

I use codex now.

Aliens playing D&D

Nah it's the simulation rendering in low poly further away. Or both.

I'll get around to reading this at some point.