mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 21:14:14 +08:00
bfae3789ba
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/51483 This PR moves the conv weights of a frozen model to MKLDNN, and AOT reorders the weights. When the weights are already in MKLDNN, just computing a single conv by converting the input and output from/to mkldnn provides large speedups. I benchmark'd the results of the top 200 shapes in predictor [here](https://www.internalfb.com/phabricator/paste/view/P171537938), as well as verified that it sped up popular models in torchvision. Test Plan: Imported from OSS Reviewed By: navahgar Differential Revision: D26696703 Pulled By: eellison fbshipit-source-id: 0b4441bee4f6e0890a4540fbca3bb5e58b8c5adf
10 lines
298 B
Python
10 lines
298 B
Python
import torch
|
|
class LinearMod(torch.nn.Linear):
|
|
def __init__(self, *args, **kwargs):
|
|
super().__init__(*args, **kwargs)
|
|
|
|
def forward(self, input):
|
|
return torch._C._nn.linear(input, self.weight, self.bias)
|
|
|
|
print(torch.jit.trace(LinearMod(20, 20), torch.rand([20, 20])).graph)
|