tysam_and

34 ms·

1. 331.▲
   by tysam_and 4y ago

   I am not from the LLM world, but I believe it's mostly constrained by the standard multiprocessing limits -- communication and synchronization of multiple workers, some of whom operate over an exceedingly slow Ethernet interface.
2. 332.▲
   by tysam_and 4y ago

   Today to me is the equivalent of the phone phreaking days when people are just doing as much as they can and getting away with as much as they can for as long as they can, until the regulations come. It will be an interesting time in the ne
3. 333.▲
   by tysam_and 4y ago

   I was so confused by the saltiness until I saw the username. I'm sure you've earned it. I got into deep learning because of your char-rnn posts a while ago -- it inspired me to do an undergrad thesis on the topic. I read arxiv pap
4. 334.▲
   by tysam_and 4y ago

   Would this be a commentary on some of the technical nature of the article? It's a rather high level paper, and seems to be part of the main fodder for Hacker News.
5. 335.▲
   by tysam_and 4y ago

   I could put some caesar dressing on this and enjoy it, but I am unsure as to the relevance to the article at hand. Could you please clarify the point you are trying to make here?
6. 336.▲
   by tysam_and 4y ago

   This is a significantly different implication by far, to the point that the two statements do not even say (or imply the same thing). The symptoms of anxiety and depression are indeed mental health issues, but changing it to the broader cat
7. 337.▲
   by tysam_and 4y ago

   To be honest, the second name has connotations that are even more terrible than the original name. They need to seek help from a better naming team, please. It would help them, us, and the world, so much.
8. 338.▲
   by tysam_and 4y ago

   This is very different, FP8 is a new and active area of research that has new hardware that is just supporting it. You are thinking of (U)INT8 quantization
9. 339.▲
   by tysam_and 4y ago

   This is very adorable. Thank you for sharing this story with us today.
10. 340.▲
    by tysam_and 4y ago

    This is a good article. I will be honest, the Twitter link felt like a very ironic finish to the article, so much so that I chuckled. Some rat races may indeed continue, though maybe it is not all or nothing. We wish the author well in his
11. 341.▲
    by tysam_and 4y ago

    For the original competition, you can check out https://dawn.cs.stanford.edu/benchmark/#cifar10 . Special note is that these are multi-GPU setups which are much more complicated and out of the reach of most average Joe
12. 342.▲
    by tysam_and 4y ago

    My apologies! Titles are hard. Indeed, this is training to 94% accuracy in only ~12.3-~12.4 seconds on a single GPU (so human-affordable), which gets us closer to massive scaled testing for neural network phenomenology. Previously a number
13. 343.▲
    Show HN: Hlb-CIFAR10 0.2.0: New world record (~<12.38s) on single-GPU CIFAR10(github.com)

    7 points by tysam_and 4y ago|5 comments
14. 344.▲
    Show HN: Hlb-CIFAR10 – 94% accuracy in 18.1 seconds, A100, ground-up hackable(github.com)

    1 points by tysam_and 4y ago|0 comments