jerpint

33 ms·

1. 571.▲
   by jerpint 4y ago

   Neat
2. 572.▲
   by jerpint 4y ago

   How is this different from dall-e mini on huggingface?
3. 573.▲
   by jerpint 4y ago

   "existence is pain!" That printer, probably
4. 574.▲
   by jerpint 4y ago

   The basic idea is to have a q,k,v cache of all the previously seen tokens that gets updated over time. The transformer can decide to do self-attention (and ignore the cache) or focus on elements from the cache (enabling it to attend to prev
5. 575.▲
   by jerpint 4y ago

   I was a little underwhelmed to see that the model was trained to imitate other SOTA RL agents, still very impressive results
6. 576.▲
   by jerpint 4y ago

   You can run HF on sagemaker, they're not necessarily competing
7. 577.▲
   by jerpint 4y ago

   I'm relatively new to Heroku (about 2 months) and haven't received a notification
8. 578.▲
   by jerpint 5y ago

   This was very entertaining. A few tells that this was imagined, apart from the obvious unlikeliness of a microwave passing a turing test so well: 1) I don't think it's possible to give 100 pages worth of prompting to GPT-3 in the
9. 579.▲
   by jerpint 5y ago

   www.jerpint.io
10. 580.▲
    by jerpint 5y ago

    Absolute wild!
11. 581.▲
    by jerpint 5y ago

    You need to run a forward pass somehow through a model, so at the very least you need something to define your operations + load your weights. You also need whatever libraries for any kind of signal processing you might do before to your in
12. 582.▲
    by jerpint 5y ago

    Yes, this is a generic workflow, but when you start adding in e.g. torch[vision, audio], scipy, or any other "useful" processing library, things get out of hand pretty quickly and you need to start being more intelligent about wha
13. 583.▲
    by jerpint 5y ago

    My frustration so far with mlops is how unnecessarily large containers need to be to serve even the smallest models. Want to serve an mnist pytorch model? It'll likely be a huge dockerfile compared to model size, exceeding the capacity
14. 584.▲
    by jerpint 5y ago

    Very impressive technical work, a relatively complex approach explained in a simple way, well done
15. 585.▲
    by jerpint 5y ago

    Very satisfying!