pavelstoev

8 ms·

1. 61.▲
   by pavelstoev 3y ago

   Happy Juniper customer since 2010. Across SRX, EX, QFX lines. What to do now, what to do...
2. 62.▲
   by pavelstoev 3y ago

   In the undergrad Control Systems course, I brute-forced Kalman Filter matrix to balance the inverted pendulum on a track experiment. Worked fine.
3. 63.▲
   by pavelstoev 3y ago

   Say no to drugs.
4. 64.▲
   by pavelstoev 3y ago

   There is still so much performance to gain by applying software-based optimizations for superior performance of transformer LLMs on GPUs. From ~ 30% GPU utilization to over 90% is possible. I wrote a white paper on this, page 5 and 6: http
5. 65.▲
   by pavelstoev 3y ago

   (while nail biting)...we are back?
6. 66.▲
   by pavelstoev 3y ago

   Looking forward to MLPerf results https://mlcommons.org/benchmarks/
7. 67.▲
   by pavelstoev 3y ago

   Try Hidet - open source deep learning compiler written in Python, generating optimized CUDA https://pypi.org/project/hidet/
8. 68.▲
   by pavelstoev 3y ago

   I agree. I am surprised that many folks, not you of course, think that is okay.
9. 69.▲
   by pavelstoev 3y ago

   Not exactly idle but only at around 30% utilization on average (measured on a ~900 GPU cluster over ~25 days)
10. 70.▲
    by pavelstoev 3y ago

    Agreed. I have 20+ Hyper-V Windows Server VMs running since 2017, seeing 400-600 Mbps live video traffic - no major crashes, only due to facility power loss or annual patching reboots.
11. 71.▲
    by pavelstoev 3y ago

    Late in 2006, as a local engineering consultant working for MBTA, I designed and then physically installed OFDM receiver station on top of the Oak Grove (Orange Line) station. Google Street View still shows the antenna present.
12. 72.▲
    by pavelstoev 3y ago

    Why do you single out Grammarly in the title of this post when other larger orgs are part of this. I have no relation to Grammarly and I appeal to your fairness. Perhaps “Large organizations making OAuth implementation mistakes” would be mo
13. 73.▲
    by pavelstoev 3y ago

    You can finetune 40B falcon on 4 x A10 with compiler optimization technology from CentML. No changes to the model.
14. 74.▲
    by pavelstoev 3y ago

    Imagine someone paid you 25c/hour for 4090 compute sharing.
15. 75.▲
    by pavelstoev 3y ago

    The claim is likely not false just like a horse pulling a buggy is several orders of magnitude faster than a snail.
16. 76.▲
    by pavelstoev 3y ago

    which is why we created Hidet, open-source deep learning compiler, written in Python and which can be easily extended to create all kinds of ML projects (wink, wink) with support for Nvidia GPUs. I am still looking for information on Mojo s
17. 77.▲
    by pavelstoev 3y ago

    I have a 12 year old and a 9 year old. They pretty much grew up during the smartphone age. We teach them to understand smartphone is not a toy but a gateway to information, knowledge and communication. When we use our personal smartphones a
18. 78.▲
    by pavelstoev 3y ago

    Humble self-promo here, may I also recommend the team at CentML who dedicated their academic life (PhD and above) to GPU optimizations for high-performance ML/AI to lower the costs.
19. 79.▲
    by pavelstoev 3y ago

    I actually very much appreciate how adult it is.
20. 80.▲
    by pavelstoev 3y ago

    To answer the question why there is not much competition, I would like to point out that the required expertise is very rare. Not specifically the chip design expertise but the expertise spanning both chip hardware and low-level software fo
21. 81.▲
    by pavelstoev 3y ago

    Thanks for your comment ! Current focus of optimizations is for Nvidia GPUs but others are in the works. Hidet comes with Hidet.Script which abstracts some of the CUDA struggles and may make the ML optimizations efforts easier to implement.
22. 82.▲
    by pavelstoev 3y ago

    Have you tired Hidet ? https://pypi.org/project/hidet/
23. 83.▲
    by pavelstoev 3y ago

    They should talk to CentML.
24. 84.▲
    by pavelstoev 3y ago

    When capacity or cost constrained (or both) consider running ML on “lesser” GPUs like A40, A10s and even consumer grade 3090/4090s. With compiler optimizations these uncover surprising price-performance ratios.
25. 85.▲
    by pavelstoev 3y ago

    Would like to see updated benchmarks supporting the claimed performance gains in this project. Currently showing torch 1.12 which is rather weak compared to the latest 2.0 and torch.compile()
26. 86.▲
    OpenAI's future plans post has been mysteriously removed(humanloop.com)

    2 points by pavelstoev 3y ago|2 comments
27. 87.▲
    by pavelstoev 3y ago

    Removed at OpenAI's request – let's discuss what this means for transparency and open discourse in AI development. Archived version of the post: https://web.archive.org/web/20230531203946/https:/&#x2
28. 88.▲
    by pavelstoev 3y ago

    This content has been removed at the request of OpenAI. Just when I went back to the post for some quote material...
29. 89.▲
    by pavelstoev 3y ago

    Unfortunately I did not know him but I will surely follow now :)
30. 90.▲
    by pavelstoev 3y ago

    I worked at RIM/BlackBerry in 2005-06. It was my first engineering job. I was in my early 20’s and at some point, way before iPhone, I went to my manager and suggested we should add a camera and some multimedia features to the device,