cgravill

directly to PTX — no DSLs, no foreign language bindings, just Rust.

Wow, glad someone is doing this, looks ideal. We’ve a cargo bike with battery failing (physical connector we can’t replace and some cells borderline), when you get to Yahama motors we’d be interested to look into it.

Is there an intention to bring the auto backend selection to the non-pip interface? I know we can configure this like you show https://docs.astral.sh/uv/guides/integration/pytorch/ but we have folks on different accelerators on Linux and remembering ‘uv sync --extra cu128’ at the right time is fragile so we just make cpu folks have the CUDA overhead too currently.

(As always, big fans of Astral’s tools. We should get on with trying pyx more seriously)

The main NuGet was problematic for a long time e.g. not providing any control over transitive dependencies (like pip at the time). You had to use https://fsprojects.github.io/Paket/ if you wanted safe and consistent resolution. NuGet since got their act together and it’s not as flawed now.

We do something similar with a mix of C++, Fortran, and CUDA dependencies. Using https://cibuildwheel.pypa.io/en/stable/ and the repair functionality to restrict the manylinux to something sensible seems to have handled the complexity - while we mostly use Debian and Ubuntu the Arch folks seemed OK too.

We want to share the wheels more smoothly though and having been looking at pyx for that. Seems promising.

It's definitely possible, and my current company were very happy to agree to 4/5 of the time for 4/5 of the salary and it's working out well for everyone. However, I've also found some organisations - that are otherwise good - to be hostile to working less than full hours. Since we work on interesting problems and they were happy to go with 80% it made it an easy task deciding which job to accept!

I should add that's it's a start-up so some weeks I work more and others less. But I still have the day my kids that I wanted.

Perhaps try quietly asking your current company? They might surprise you and start a trend.

Some comparisons of power generation by deaths here: https://ourworldindata.org/safest-sources-of-energy

There are some companies such as https://www.owlstonemedical.com/ who’ve made some progress on medical diagnosis from breath

Link here to the comment https://twitter.com/richturn_ms/status/1245481405947076610

There's a related Microsoft Research project https://www.microsoft.com/en-us/research/project/project-sil...

(I work in the same lab but on different things)

Hibermate are very good, I use them especially when travelling. Original style are reasonably priced https://www.hibermate.com/products/the-original-hibermate

There's a Microsoft Research project that does something like this with mosquitoes for disease management: https://www.microsoft.com/en-us/research/project/project-pre...

(note: I work at MSR, but on other projects)

It'd be great to have it built-in but you can get https://chocolatey.org and MinGW pretty easily:

  iwr https://chocolatey.org/install.ps1 -UseBasicParsing | iex

restart prompt, then:

  choco install mingw

(no machine reboots, disclaimer: I work at MSR, nothing related to this)

I use WebSharper at MSR and have been very happy with the results. We have substantial amounts of F# for simulation, modelling, and analysis of biological systems. It’s useful to be able to run this code in many places and WebSharper allows us to transpile into JavaScript for running directly in a browser with only minor changes to the core code.

Disclosure: I work at Microsoft Research but not on F# and have ongoing interactions with folks at Intellifactory

There are some timing comparisons of using PyPy vs CPython with nltk showing improvements

http://groups.google.com/group/nltk-dev/browse_thread/thread...

The cost of providing power to datacenters is going up both in absolute and percentage. You pay once to heat the server in waste energy then again for air conditioning to cool them back down. The ARM chips provide better perf per watt but the upper limit per ARM chip is way lower. To use them, it requires distributed software which is hard but we're getting there.

"Standing desks and chaise longues are good options. Ball chairs, which bounce your spine into a C-shape, are not."