Tpt

My bet (I don't think there is any confirmation of it) is that the AMD board is still the one released last year. The main difference I see is that the new Intel board uses LPCAMM2 memory whereas the AMD board relies on usual socketed memory that has higher latency and is more memory hungry.

I guess it's mostly because cpython has a fairly good C API allowing PyO3 to just write "safe" wrappers on top of cpython APIs and provide macros to generate boilerplate whereas wasm-bindgen has to generate both Rust and JS sides and deal with the painful linear memory intermediate.

Igalia is a quite specific "consulting" agency that employs developers and get contracts from client to implement specific features or fix specific bugs, usually around FOSS. They have people who knows how to contribute to Firefox, Chrome, WebKit, Linux, Mesa... It's the go to company if you want to get something done in these projects when not having the resources for that in house.

For example they work for Valve to make the Radeon drivers better and got a grant to get basic MathML support done in the three major web browsers.

It's amazing to see MathML moving forward. As a european, I really like this usage of my taxes euros.

UV for packaging and dependency management, Ruff for linting, Mypy for type checking (will be likely replaced by Ty when ready) and whatever editor people like (PyCharm, VSCode, Helix...)

A lot of large castle owners in France have setup a small apartment on the side for winter and use the main rooms only occasionally and during summer. This is not something new, in Versailles you can visit the king actual private bedroom that is much smaller and easy to heat than the official one.

There is slightly more information on Helsing website: https://helsing.ai/centaur

No, it's the first day of the holy week

There is now libgccjit that aims at allowing to embed gcc https://gcc.gnu.org/onlinedocs/jit/

There is an alternative backend to rustc that relies on it.

If I understand correctly, this library provides some Torch kernels customized for AMD hardware. Why haven't they just upstreamed them to PyTorch for better adoption? Also, they seem to demo usage with Torch default eager execution mode and not Torch JIT/TorchScript. Is this library compatible with TorchScript?

But before becoming priest there is a quite long training program (often 5-6 years). Dropping out of it is quite common.

And even after becoming a priest, they can asked to be relieved for their obligations, including celibacy. But in this case they would have to find an other job.

Rail lines are massively subsidized and the French government is regularly bailing out SNCF (the national rail company). So imho it not clear that the global market distortion is in the favor of plane.

Location: France

Remote: Yes

Willing to relocate: No

Technologies: Rust, Python, PHP, JS, Lua, Java, SQL, SPARQL, PyTorch, MediaWiki, NLP, Linked Data...

Résumé/CV: https://thomas.pellissier-tanon.fr/cv.pdf

Email: thomas at pellissier-tanon dot fr

I am a software engineer with a strong research background. I am mainly focusing on data management, databases, and NLP.

See my website for a more detailed description: https://thomas.pellissier-tanon.fr/

There are nice libraries for that like https://ontop-vkg.org/ that translate sparql queries to sql queries and provides some OWL inference.

the customer can run, patch, and run as patched, but not share with others. the seller guarantees the software will work as documented, won’t be infected with malware, won’t be riddled with security holes, won’t contain plagiarized code.

The two points together are dangerous: they make the seller commit that the software works as intended, even if it is patched by the customer.

The second point is also terrible for the seller: it creates an unbounded liability in case of bug or security issues. Let's assumme that a security vulnerability is discovered, the seller is liable for its exploitation even if they released a patched version that the customer has not applied.

I am currently building a single-node graph database implementing the SPARQL query language. It is written in Rust and is named Oxigraph: https://github.com/oxigraph/oxigraph

I would love to get help, I have a working MVP but there are a lot of work left to get good performances for which I would love help. Some current problems are:

- Very big write amplification (the database size on disk is very large).

- The query optimizer is basically non existent.

- The query evaluation code might be easily optimized further.

There is not yet a lot of developer-focused documentation. Feel free reach out here or on Gitter: https://gitter.im/oxigraph/community

Apple is maintaining a Tensorflow plugin: https://developer.apple.com/metal/tensorflow-plugin/

The neural engine seems to be only about inference. For training it seems most systems rely on Metal like the Apple Tensorflow plugin [1]. But I have never tried to do ML on macs so I am maybe wrong.

[1] https://developer.apple.com/metal/tensorflow-plugin/

I'm very curious to see how the 64GB of RAM GPU performs with deep learning models fine-tuning. It might be impressive.

I like sled that is a nice embedded key value store written in Rust: https://sled.rs/

However, it is still in heavy development and a bit of a moving target even if the developers are currently heading toward stabilization of the file format.

In the RDF world, there is Redland: https://librdf.org/

But I don't know if it scales up to Wikidata size.

GitHub pages. It's free, you just have to push to a git repository to update it and it allows custom domain names.

There is also Solid from Tim Berners-Lee https://solidproject.org/

Yes, but for reducing ops and maintenance costs by using new techs you need people to mamke changes to the system. And it's probably not something volunteers would want to commit to do.

I don't think they where ever Wikimedia Foundation staff paid to work on Wiktionary, Wikibooks, Wikiquote, Wikivoyage, Wikisource Wikispecies, Wikinews and Wikiversity. So, these projects cost to the Wikimedia Foundation the hosting (probably around ~100k/year at most) and some other small cost (trademark, some community support). Wikidata is developped by Wikimedia Deutchland that is an other organization than the Wikimedia Foundation and has its own fundraising. So, I believe that if you donate to the Wikimedia Foundation nearly all your donation is going to Wikipedia.

You probably would like to have a look at Wikidata that is very similar to Freebase and is gaining a huge community of contributors and data reusers: https://wwww.wikidata.org/

Yeah. But try going anywhere around Lyon[0] after the nice TGV trip from Paris[0] and welcome to connection hell.

Public transportations are often quite good in European cities. When I was in college, it tooked me sligtly more than 3 hours door to door to go from my parents home in a Paris suburb to my University dorm in Lyon. It's not so bad for a trip of nearly 450km.

Also what's the difference between WikiData and DBPedia?

Wikidata is a Wikimedia project with aim to create a structured knowledge based. It is mostly filled and curated by humans: https://www.wikidata.org

DBPedia is a knowledge base which content is extracted from Wikipedia (mostly from the infoboxes). It is a project run by researchers: http://dbpedia.org

You should have a look at [1] that outputs an HTML rendering of pages with a lot of metadata.

[1] https://en.wikipedia.org/api/rest_v1/#!/Page_content/get_pag...

There is https://github.com/livingbio/syntaxnet_wrapper that does the job fairly well (I also spent days trying to be able to pass to SyntaxNet different textes without having to reload the model). Warning: installation is a bit difficult.