apaszke

5 ms·

1.
Mlir-hs – Haskell bindings for MLIR(github.com)
3 points by apaszke 5y ago|0 comments
2.
by apaszke 9y ago
PyTorch was never a Chainer fork. The whole codebase are C libs from Lua Torch, and a bunch of Python code that was written entirely for this project. Chainer was an inspiration, but no code was ever shared between those two projects.
3.
by apaszke 9y ago
PyTorch uses NVIDIA NCCL for multi-GPU communication (under BSD license). Gloo is only one of the three backends that can be used for distributed communication out of the box.
4.
by apaszke 9y ago
It's not controlled by Facebook in any way. It's true that a large part of the core team works there, but development is public and guided by community needs first.
5.
by apaszke 9y ago
It's not like you have to give up a lot - the graphs are simple data structures and creating them is not the expensive part of the training. The computation has to be re-done at every step in a static framework too, and this is the par
6.
by apaszke 9y ago
gloo is only one of the three currently supported backends. One can easily switch to MPI, and pick an implementation that comes with a license you want.
7.
by apaszke 10y ago
It supports numpy, and conversion from PyTorch tensors to numpy arrays is a matter of calling the numpy() method.
8.
by apaszke 10y ago
To not have the kernel launch overhead you'd need to stop launching GPU kernels but that's now how things work in any framework ;)
9.
by apaszke 10y ago
These ops are just not needed in PyTorch. while is just a Python while loop. Scan is a for loop, map is a list comprehension that applies modules. No need for anything fancy.
10.
by apaszke 10y ago
Yes! There's an issue on that, where we'll be coordinating the work: https://github.com/pytorch/pytorch/issues/494
11.
by apaszke 10y ago
You save the parameters and the code of the model definition
12.
by apaszke 10y ago
torch?
13.
by apaszke 10y ago
But it is possible to write your model in purely functional style. Check out the PR to examples repo with functional ResNets https://github.com/pytorch/examples/pull/22 .
14.
by apaszke 10y ago
torch.nn offers a very similar interface to Keras (e.g. see Alexnet definition at https://github.com/pytorch/vision/blob/master/torchvision/mo... ).
15.
by apaszke 10y ago
No! Python 0 based indexing everywhere.
16.
by apaszke 10y ago
The core libraries are the same as in Lua torch, but the interface is redesigned and new.
17.
A quick tour of Torch internals(apaszke.github.io)
1 points by apaszke 11y ago|0 comments
18.
by apaszke 11y ago
I am aware of 'ul' conflict, but I haven't heard of it before, so I just hoped that it's not popular :x If you've found a bug in installation script then please post an issue at GitHub :)
19.
by apaszke 11y ago
I am aware that there are lots of universal window managers, but I just wanted something simple that would have a terminal API. This is why I've created termtile
20.
by apaszke 11y ago
I know that there are loads of universal window managers, but I wanted a one with a terminal API ;)
21.
by apaszke 11y ago
Thanks! It's nice to hear that! What is the bug you're talking about?
22.
by apaszke 11y ago
I didn't include it, because OS X can already make the window full screen. Do you think it would be useful just to resize the window without creating a new desktop?
23.
by apaszke 11y ago
Hey guys, I'm a creator of termitile. Many of you have pointed out that there are lots of window managers for OS X and my intention wasn't to create just another one. What I really needed was a bunch of script that I could alias a
24.
by apaszke 11y ago
I know about it, but I didn't have an 'll' alias so i picked it just to be consistent with 'rr'. If you wish you can configure the aliases in the install.sh file (if you decline to overwrite an existing alias it ask