aaronblohowiak

>What I want to see is comparisons to NON-quantized models

isnt that the first image in the diagram / the 22GB model that took 111 seconds?

>I'm curious how one ends up with such ahistorical sequence.

they are all actively in-use.. if gp is earlier in their career, it could all be in last 10 years.

Yea all of that is stuff I do not want to bother with. At all. “It’s nice if you take on maintenance burden of a bunch of additional moving parts” is the opposite of what I want. If you have to support a diverse set of languages / runtimes / environments and you deploy using containers, maybe it makes sense, but that seems like a use of the complexity budget I’d rather spend on… something else

TMTOWTDI is antithetical to python as a language it’s so odd to me that its packaging and env management is the opposite.

Do you have a suggested solution? Im solo dev for now but will be adding more folks in foreseeable future. Stuck with python for some things (notebooks, model development.)

Like PLCs or something similar?

Many municipalities near here have restrictions on days of use. It’s by neighborhood so you have only one really bad day a week but lawn services can still operate around the week.

The hoops that people go through to solve this sometimes creates something even more complex and not great, like forcing all development into a docker container..

Ever try conda though? I’ve had moderate success with pipenv, but tbh I don’t love it as it hides too many things when installing a package fails.

It uses technological differences as key plot and setting components not just space as sea, so it is sci fi but it is improbable in many ways so yea “soft” sci fi or more speculative fiction

I like this. Couple things to add:

Fast setup and revision are important but incomplete list of maintenance tasks; are metrics/logs predictably named and accessed? Can you perform manual experimentation without hard-to-configure client (ir: hit the server with a browser or run a cli)?

Also, "cycle time" or "revision time" are soo important, but I havent found a good way to do that with AI model development :( any tips here?

not sure what you mean, but htmx might be what you are looking for?

Kinda related, hopefully someone here in comments can help: what’s your favorite precise nn search that works on arm Macs for in memory dataset; 100k times / 300 float32 dims per item ? Ideally supporting cosine similarity

Faiss seems big to get going, tried n2 but doesn’t seem to want to install via pip.. if anyone has a go-to I’d be grateful. Thanks.

Central banks will not tolerate deflation, too destabilizing

Steel is sooooooooooooo much harder than al, it’s not even in the same ballpark. Aluminum is about as hard to cut as wood. For non round holes, you are looking at approximations or broaching. Look up manual hex broaching

What about the lumen from opulo?

Which framework? Do you write sync or async? I’ve AoC’d rust and really liked it but async seems a bit much.

>or why anyone would reach for Python for a webapp in 2024

I'm out of touch.. why not?

for other folks wonder what the acronym means; RF in this context is Random Forest

>they get enough idioms in C++ slightly incorrect

this is part of why I stay in python when doing ai-assisted programming; there's so much training information out there for python and I _generally_ don't care about if its slightly off-idiom, its still probably fine.

can I get a lidar for 1k like the cars use?

in concrete, cement makes up less than 25% of the mix, and here corn is like 20%, so I think when you're comparing this class of materials (concrete, epoxy granite, concrete), it makes sense to describe in terms of the binder even if it is a smaller % of the complete product.

Like a thermal printer in general or specifically attached to a calculator?

In general, an automated system to catalog, store and retrieve trading cards (magic for now, Pokémon next.) Right now, working on computer vision recognition of magic cards, but I’m trying to tackle the hard edge cases that many other projects and products don’t (revised vs unlimited, multiple reprints in The List, automatic language detection, etc etc.)

it is planned for decommission. in the interim, they are isolating the impacted module and not planning repair (at least, that's what I got from the article!)

this is not true. https://wiki.bambulab.com/en/knowledge-sharing/enable-lan-mo...

So you embed magnetic particles in silicon rubber and magnetize them, then use magnometers to detect how the magnetic field is changing from a few different points of reference in order to detect the deformation of the rubber and from that you can analyze the "pressure points" on the surface. the innovation here is that you dont have a lengthy re-calibration of your "input signal" to the particular magnet-infused silicone interface because the manufacturing makes them consistent enough to be replaceable parts?

this makes advanced touch sensors more like machine-cut screws than bespoke hand-forged nails.

I have good news for you, it gets worse! Look at very large and old ruby codebases :)

At what feature size? http://sam.zeloof.xyz/category/semiconductor/

They are leaders in (very long) HVDC and UHVDC which helps bring energy from west to east, an alternative to building lots of time-shifting capacity.