Matthew Honnibal

@honnibal

Computational linguist from Sydney and Berlin. 💫 Author of the NLP tools. 💥 Founder

Vrijeme pridruživanja: svibanj 2008.

Medijski sadržaj

  1. 17. stu 2019.

    Had a great time in Chennai for – thanks for having me! I spoke about our ongoing spacy-transformers work, especially design questions for longer pipelines. 📄 Slides: 🛸 spacy-transformers:

  2. 12. lis 2019.

    It's starting! giving her keynote at PyCon India here in Chennai. Slides and hopefully recording will be up shortly

  3. 2. ruj 2019.
    Odgovor korisniku/ci
  4. 31. kol 2019.

    Okay revision 3: * Get blower-style GPUs * Revert changes to the PSU (worried 1000 watt is too small, and can't find any 1500 watt Platinum in stock). * Use the CPU fan from the original build, to make sure it doesn't block the GPU.

    Prikaži ovu nit
  5. 31. kol 2019.

    Thinking maybe I should give up and get something prebuilt. This water-cooled dual 2080 TI build probably performs great. Can I get it without the 🚨, or would that be an affront to the artist? Computer science 2019: all extremely normal 🙃

    Prikaži ovu nit
  6. 30. kol 2019.

    After lots of great feedback, I've got an updated parts list for my deep learning workstation. I took the advice to add a second GPU, more RAM, and extra storage. Also thanks to for the advice on cooling and the PSU. Any changes I should make before I go get all this?

    Prikaži ovu nit
  7. 28. kol 2019.

    Finally getting around to buying a deep learning workstation. Current draft parts list: , based on 's guide. I really don't have much experience at this, so: anything I should change?

    Prikaži ovu nit
  8. 9. kol 2019.
    Odgovor korisnicima i sljedećem broju korisnika:

    It depends what you're looking to discover! For spaCy I've actually had to do quite a bit of re-architecting to get the models efficient on CPU. Our NER model is pretty different from others: .

  9. 28. lip 2019.

    Have been working on spaCy cloud computing to prepare materials for the training days next week. is very cool: now parsing common crawl text at about one million words per second (for about $2 per hour on GCE)

    Prikaži ovu nit
  10. 21. tra 2019.

    Today's off-beat use of : Training myself (for German practice), instead of a machine learning model. I wanted to script my own spaced repetition prompts, and Anki was really frustrating. It's nice being able to queue up the questions from Python.

    Prikaži ovu nit
  11. 19. ožu 2019.

    Nice to see our hard work on removing subtle system dependencies from spaCy paying off already. Almost 20x faster performance on Google App Engine between v2.0.18 and v2.1.0 😎

    Prikaži ovu nit
  12. 27. sij 2019.

    Hard at work in Sydney.

  13. 6. sij 2019.
    Odgovor korisnicima

    Admittedly reading the code to get the model architecture's a bit hard. I haven't written this up because I've still been changing things. The best reference are these videos:

  14. 2. sij 2019.

    Was it the fame that brought Knut down? Maybe, but maybe not --- because there's a final twist in the tale. You see, Ernst got famous eventually too, but in a completely different way. You might recognise this picture of him.

    Prikaži ovu nit
  15. 2. sij 2019.

    Ernst was a sun bear raised by a loving mother. Amongst all the Knut-mania he was often forgotten, so he became something like the indie pick. Yes yes Knut is cute -- but did you see Ernst as well? For a while Ernst and Knut were even roommates, sharing an enclosure.

    Prikaži ovu nit
  16. 2. sij 2019.

    Knut-mania continued with many memes. Personally, I always liked the song the best: . But while all this was going on, Berlin Zoo had *another* baby bear that was no less cute!

    Prikaži ovu nit
  17. 2. sij 2019.

    Thus Knut-mania was born. 400 journalists visited the zoo on "Knut Day", when he was first presented to the public. He joined Leonardo DiCaprio on the cover of Vanity Fair. Azerbaijan made a commemorative stamp.

    Prikaži ovu nit
  18. 2. sij 2019.

    Albrecht lost the sloth bear case, so when Knut was orphaned, he wanted to point out that this crazy law said this bear must be euthenised as well. Alas, tabloids gonna tabloid. Bild's take on the story was basically ANIMAL RIGHTS ACTIVIST WANTS CUTE BEAR MURDERED.

    Prikaži ovu nit
  19. 22. stu 2018.

    Starting soon (1h 45m): will explain why ✊The AI Revolution Will Not Be Monopolized at . Live stream here:

    Prikaži ovu nit
  20. 2. kol 2018.

    In NLP and ML we talk a lot about models and optimization. But this isn't where the battle is really won! I've been trying to explain my thoughts on this lately. Big thanks to for a great event. 📺 Slides:

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·