This doesn't mention the transformer models [1] that were recently announced, so even though it's not 2.2 related, I'll just point it out because it's recent spaCy news :)
I had thought I'd seen somewhere (spaCy IRL?) that this kinda stuff was not supposed to show up until v3.
We're definitely planning to do more there -- we took a bit longer than anticipated getting v2.2 out, due to various packaging and automation hiccups.
Huggingface are doing a great job keeping transformers moving quickly, so we already have some catching up to do. We need to update the wrapper so it works with the new version. I also want to check out Tensorflow 2, and build a Thinc wrapper for it.
Definitely a fair concern, and thanks for the kind words :).
We haven't had to fund-raise, so our company performance has actually been a bit under-the-radar. But I can assure you everything's been going well, and most people would be surprised at the numbers. We've been profitable since day one, but this year we've been making enough to hire three more full-time developers: https://explosion.ai/about .
Our revenue all comes from our annotation tool Prodigy: https://prodi.gy . Prodigy's very popular, we have a few thousand users in total. There aren't many alternative tools that are scriptable like Prodigy is, and annotation is something that has to be done on almost every project. Even if you don't need to annotate training data (e.g. if you're training on click data or something), you'll usually want to do small annotation tasks to do error analysis and understand your performance.
Nice! I'm excited! I've been working on a heavy NLP project with spanish and been having some issues and so this will be nice to test out and see if it helps!
9 comments
[ 3.6 ms ] story [ 27.1 ms ] threadThis doesn't mention the transformer models [1] that were recently announced, so even though it's not 2.2 related, I'll just point it out because it's recent spaCy news :)
I had thought I'd seen somewhere (spaCy IRL?) that this kinda stuff was not supposed to show up until v3.
[1] https://explosion.ai/blog/spacy-pytorch-transformers
So probably not directly related to core SpaCy itself.
Huggingface are doing a great job keeping transformers moving quickly, so we already have some catching up to do. We need to update the wrapper so it works with the new version. I also want to check out Tensorflow 2, and build a Thinc wrapper for it.
https://www.urbandictionary.com/define.php?term=spacky
Although I understood it to be derived from:
https://www.urbandictionary.com/define.php?term=spacker
We haven't had to fund-raise, so our company performance has actually been a bit under-the-radar. But I can assure you everything's been going well, and most people would be surprised at the numbers. We've been profitable since day one, but this year we've been making enough to hire three more full-time developers: https://explosion.ai/about .
Our revenue all comes from our annotation tool Prodigy: https://prodi.gy . Prodigy's very popular, we have a few thousand users in total. There aren't many alternative tools that are scriptable like Prodigy is, and annotation is something that has to be done on almost every project. Even if you don't need to annotate training data (e.g. if you're training on click data or something), you'll usually want to do small annotation tasks to do error analysis and understand your performance.
I love spaCy and how its improving all the time