liuliu

62 ms·

841.
by liuliu 7y ago
I've come back from Ruby as the default scripting language to Python. Over the years, the accumulated Python scripts are much more consistent and easier to read than aged Ruby script. Ease of use, sure. But the language itself I would
842.
by liuliu 7y ago
Precision reduced to 2-bits, with "codebook" of 2 within each 4x4 patch. Similar to https://en.wikipedia.org/wiki/Block_Truncation_Coding but use 1 more bit. There are a lot of tuning left to do to find bette
843.
Reducing Memory Usage for DNN Training with Tensor Compression(liuliu.me)
25 points by liuliu 7y ago|3 comments
844.
by liuliu 7y ago
Yes, maybe. But it is at least unclear whether is inherently CNN or not (many people would vote for no). ImageNet training data is heavily texture dependent, therefore, the trained model leaning towards texture information more. Some recent
845.
by liuliu 7y ago
Not really. We got lucky last time. It is not given. Given the advantages the U.S. has over nuclear weapons, there is a big asymmetry here (not to say symmetry is more stable, but to layman's eye it intuitively feels so).
846.
by liuliu 7y ago
New (or kinda new?) Framer X tries to do that for React: https://www.framer.com/development/
847.
by liuliu 7y ago
Maintaining a pipeline that works takes effort that can be spent somewhere else. As long as it can be reproduced with a little effort from me, I really don't mind. I would rather they spend time to try a lot of different things, build
848.
by liuliu 7y ago
Unlikely though. TR is AMD's only HEDT presence (qual-channel, high core count). X570 chipset will likely s be dual-channel memory, limiting to 64GiB, seems still positioned for mainstream. Intel currently have very high price HEDT off
849.
by liuliu 7y ago
I am still not sure what they meant "client systems on shelf for 2019". All the other sources seem to imply it strictly for mobile (Ultrabook and NUC). So the server 10nm will be first (H1 2020) and then desktop later than that? D
850.
by liuliu 8y ago
My understanding is that Intel is doing OK with smaller die size 10nm (Cannon Lake-U / Ice Lake-U) for Ultrabooks. But for bigger dies such as HEDT or server CPUs, the Ice Lake SP is still pretty much in the air with Intel's vague
851.
MDAU, DAU and the Growth Hacks for High-Growth Companies(drz.today)
1 points by liuliu 8y ago|0 comments
852.
by liuliu 8y ago
https://drz.today/2019/mdau-dau-and-the-growth-hacks-for-hig...
853.
by liuliu 8y ago
Can you expand a little bit? I often find if I skip one or more steps mentioned here, the later debugging is tremendously harder (and often involves go back to these steps again). Some of these advice like visualization are well supported i
854.
by liuliu 8y ago
https://www.bizjournals.com/portland/news/2017/02/10/intel-c... Not sure if that is still true after Krzanich out.
855.
by liuliu 8y ago
Yeah, it is interesting to see whether Jason Evans is interested enough to port some of the ideas over to jemalloc. My original wording is poor and shouldn't be read as suggesting merging the codebase literally. Just from reading the p
856.
by liuliu 8y ago
On free, other allocators are likely to return the last freed address but with mesh it will return a random location (i.e. hurt temporal locality). But I guess in modern CPUs it doesn't matter that much because that whole page probably
857.
by liuliu 8y ago
I think there are dozens of inference frameworks. But full featured ones probably just TF / PyTorch / mxnet / Chainer that I am aware of (since caffe2 is merged into PyTorch).
858.
by liuliu 8y ago
This is the first time I seriously read the paper and the code. Does this mean for ODE net, we are sharing weights for f? (even though we evaluate f at different y and t points)? If that is the case, seems you can implement a full static ve
859.
by liuliu 8y ago
As a by-side casual observer, if the military continue to get funded while civilian departments continue furlough for months if not years, would that creates some weird social dynamics?
860.
by liuliu 8y ago
Yeah. Frequency is higher, but not sure if you can overclock it. The biggest problem I saw on these EPYC chips (potentially) is that the PCIe lanes spread across 4 dies. So the performance of data going through host would be interesting to
861.
by liuliu 8y ago
No. Two slots away is plenty.
862.
by liuliu 8y ago
For RTX cards, water cooling probably is a must nowadays. Had a lot of cooling issues with my 3 2080Ti FE. Although managed to get it stablize at 85C, it is not ideal. Also, getting as much as PCIe lanes is not a bad idea. The RTX cards onl
863.
by liuliu 8y ago
I am genuinely curious what's out-of-line mean here, especially comparing with Doxygen. Doxygen can generate xml output, and these xml output are easy to parse to either generate markdown, or rst, and then feeding to to Jekyll or Sphin
864.
by liuliu 8y ago
I switched to Syncthing with the main desktop backed up to S3 with zbackup (encrypted). Syncthing works really well (and faster!) when keeping laptops and desktops in sync for my case.
865.
by liuliu 8y ago
Before deep learning is a thing, most computer vision tasks are done either in greyscale, or in a more reasonable colorspace (HSL, YUV etc.). Thus, you most likely just call cvCvtColor the first thing when extracted the frame, and it doesn&
866.
Coroutine to Coordinate CUDA Streams(liuliu.me)
2 points by liuliu 8y ago|0 comments
867.
by liuliu 8y ago
I don't think this is possible. At least in TensorFlow 1.x, a lot of non-trivial parts (other than Operators and Gradients) are implemented in Python directly. See their Java bindings, the functionalities exposed are miles far from the
868.
by liuliu 8y ago
My reading is that they didn't look at the Linux implementation (as "clean-room").
869.
by liuliu 8y ago
Will try not to reply here since it can be inflammatory (any cryptocurrency topic here HN). That's been said, can you find any mention of "messages" in this article: https://steemit.com/dpos/@dantheman&#x
870.
by liuliu 8y ago
The PBFT algorithm, or any Byzantine fault tolerant algorithms are deliberate on what's the message-passing scheme and election protocol would look like. However, in recent ICO craze, a lot of algorithms (with the exception of Bitcoin,

More ›