tardyp

LLM Model versioning really makes me perplex those days...

New maintainers never bothered to change the range.. History num2words is based on an old library, pynum2word, created by Taro Ogawa in 2003. Unfortunately, the library stopped being maintained and the author can't be reached. There was another developer, Marius Grigaitis, who in 2011 added Lithuanian support, but didn't take over maintenance of the project.

I am thus basing myself on Marius Grigaitis' improvements and re-publishing pynum2word as num2words.

Virgil Dupras, Savoir-faire Linux

Interesting to see ppl caring about marshalling overhead when working with LLMs

A shame it doesn't use udsoncan as a backend. Udsoncan is pretty well done and indirectly support doip which is going to be ubiquitous in those new software defined vehicles..

MIT is not a gift. It still requires attribution, notice and license term.

This lawsuit is even not about copy left.

Another key reason to me that open source can be very good quality is when the developer is the key user.

If the developer has strong knowledge of the business while designing a product, it always ends up with better qa as they are writing the right product on first draft.

You hardly get that when you hire someone to do something for you. That is valid in other domain than programming.

cool. It was hard to find. :)

looks like only a100 accelerator is documented.

cannot find any info about P4, T4 or P100

Very helpful! Thanks!

What could be missing is the GPU aspect. Not sure if the data is available, but it is very unclear to me which GPU is available on which machine type. and also how many, so that I have some chance to get one.

Wordle will become the new todolist

Looks like QuickTransit was a jit engine which was the base of Rosetta 1. Rosetta 2 is AOT translation.

It looks like websocket+msgpack would have similar overhead once the initial initial overheard is passed.

But you won't miss all the http reverse proxy ecosystem.

Good ol times when front page videos were 40k views max...

Can you elaborate? This does not make sense to me

Python ambee is a simple client library for a proprietary sass iot service. I guess they want to control exactly how their users are using their service.

I wonder why they didn't just make two separate routines without any branch except for loop. Then you would just make another safety offset to assume for the last loop iteration prediction.