Timothy Liu

@timothy_lkh_

Deep Learning, NLP. I love GPUs. Opinions are my own.

Singapore
Vrijeme pridruživanja: srpanj 2011.

Tweetovi

Blokirali ste korisnika/cu @timothy_lkh_

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @timothy_lkh_

  1. proslijedio/la je Tweet

    🌩️ It's not just PC gaming in the cloud. It's GeForce gaming in the cloud. Give yourself the with GeForce NOW — anywhere, any device, on demand. The wait is over. Available now. Learn more →

    Poništi
  2. proslijedio/la je Tweet
    4. velj
    Poništi
  3. proslijedio/la je Tweet
    4. velj
    Odgovor korisniku/ci

    No love for ? ask boss come support our students. In SUTD we work 25h/day to create a better world by design. Losing housing to make room for LOA students was understandable but devastating cover our story

    Poništi
  4. proslijedio/la je Tweet
    3. velj

    In January, , , and I ran a short class at on topics we think are missing in most CS programs — tools we use every day that everyone should know, like bash, git, vim, and tmux. And now the lecture notes and videos are online!

    Poništi
  5. proslijedio/la je Tweet
    2. velj

    The full set of my 2019 graduate-level computer architecture course lectures at ETH Zurich is online, along with all lecture videos, slides, and course materials: Course schedule: Youtube playlist: First…

    Poništi
  6. proslijedio/la je Tweet
    2. velj

    Wow: Google's "Meena" chatbot was trained on a full TPUv3 pod (2048 TPU cores) for **30 full days** - That's more than $1,400,000 of compute time to train this chatbot model. (! 100+ petaflops of sustained compute !)

    Poništi
  7. proslijedio/la je Tweet
    2. velj
    Odgovor korisnicima

    TF and PyTorch are two that are not going to be solved by HIP. cuDNN has a big head start and I don’t see anyone being able to get performance parity without a different approach (tensor compiler like MLIR). Big project. My guess is they punt and backfill with field eng.

    Poništi
  8. proslijedio/la je Tweet
    31. sij

    Learn to build an interactive Transformer attention visualization based on and in under 30 minutes! We developed a minimal teaching example for our IAP class, publicly available here:

    Prikaži ovu nit
    Poništi
  9. proslijedio/la je Tweet
    31. sij

    . is the only place crazy enough to conceive of rendering 4K 60+ Hz graphics at home by running a neural network for every frame. DLSS: Trained in , running on the Tensor Cores in RTX GPUs, redefining graphics with AI. And we’re just getting started!

    Poništi
  10. proslijedio/la je Tweet
    29. sij

    this is mesmerising (brickbrosproductions on insta)

    Poništi
  11. proslijedio/la je Tweet
    27. sij

    Tired of looking at your logs in silos? Increase the quantity and variety of logs analyzed with . In our blog, we analyze +300k raw alerts in under 3 seconds, including co-occurence analysis and rolling timeseries of alerts -

    Poništi
  12. proslijedio/la je Tweet
    25. sij

    experienced pretty much the same and i bet others did too. don’t get fooled with the reduced Flops! the speed of depthwise conv can be disappointing!

    Poništi
  13. 25. sij

    Updated my blog post on the performance of depth-wise separable convolution, with additional analysis from profiling the GPU kernels and comparisons between GPU, CPU and TPU.

    Poništi
  14. proslijedio/la je Tweet
    24. sij

    Pascal's paper on the "expected gradient" is published:

    Poništi
  15. proslijedio/la je Tweet
    Odgovor korisnicima

    So far, this seemed to give the best context (from a Pulitzer Prize winner who covered SARS):

    Poništi
  16. proslijedio/la je Tweet

    🧧Team wishes our community friends a very Happy Lunar New Year! May good health, good luck, and happiness fill your home throughout the year. 🧧

    Poništi
  17. proslijedio/la je Tweet
    24. sij

    Interesting analysis suggesting that the reason for the disappointing performance of many modern CNN architectures is that their depthwise convolutions are memory-bound.

    Poništi
  18. proslijedio/la je Tweet
    24. sij

    Scaling Laws for Neural Language Models. OpenAI team found that the loss of LM scales as a power-law with model size, dataset size, and the amount of compute used for training up to seven order of magnitudes.

    Prikaži ovu nit
    Poništi
  19. proslijedio/la je Tweet
    23. sij

    The NVIDIA T4 is now more accessible than ever before with . Offering a broad selection of accelerated compute for every workload, performance level, and price point. See how GPUs can accelerate everything from deployment to 3D visualization.

    Poništi
  20. proslijedio/la je Tweet
    22. sij

    My blogpost on how & why we use convolutional neural networks as a model of the visual system is probably the most read thing I've ever written and it's now been expanded & updated into a proper review article, complete with 136 references & 5 new figures!

    Poništi

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:

    ·