-
Convolutions provided by the [`FastConv` package](https://github.com/aamini/FastConv.jl)
Described in [their paper](https://arxiv.org/pdf/1612.08825.pdf) is considerably outperforming the back ends…
-
### Description
Transpose convolutions are orders of magnitude slower than their complementary regular convolutions and their counterparts in torch (at least for the sizes in the example below). Thi…
-
Hi Phil, I tested it in my private project 2 days ago, and it seems to speed up learning quite significantly, not sure that final val/train losses are better, more like very similar to original but it…
-
### 🐛 Describe the bug
On running torchaudio code I noticed that some resampling operations are slower than they should be on the forward pass of the Resample transform. I tracked the slowness to th…
-
Implement Fast Fourier Transformations for convolutions
-
The benchmarks this time around are interesting, with some fairly clear trends emerging for the near future.
### Looking Back
First, some appreciation for where things are,
- 9 months ago, we were ~3…
-
It would be great if we had a direct convolution kernel, which would probably be faster for small convolutions.
-
From https://github.com/pytorch/pytorch/pull/134282#issuecomment-2307157197, in the aarch64 dashboard results, if we benchmark with fp16, it is 2x~10x slower than bf16, often causing timeout in cases.…
-
您好。打扰了。我想问下AssembledBlock是您自己浮现的还是AsConvSR: Fast and Lightweight Super-Resolution Network with Assembled Convolutions他们官方的代码?
-
Hi,
I've noticed that my BrainScript network trains **much** slower with version 2.4 than with version 2.3.
In CNTK 2.3 it trains with 25.4 samples/s and in CNTK 2.4 only with 11.7 samples/s. In P…