Installing a production line version of Pytorch

deployment
1

Dear to whom it may concern,

I have tried to install Pytorch on my instance but found that it is too big of a file (at ~700MB). Is there a minimal version of Pytorch (smaller size) that I could install for the main purpose of functioning in the production line (requires a few functions)?

Below are the only imports that were used:

import torch
import torch.nn as nn
from torch.distributions import MultivariateNormal
from torch.distributions import Categorical

If there is, may I have a recommended approach to install the CPU and/or the GPU supported versions of Pytorch (with terminal lines)?

Thank you so much!

2

No, unfortunately there are no pre-built binaries with a subset of PyTorch (at least no official ones) and I’m also unaware of building a subset from source.

I don’t know which methods you are exactly using but import torch would import all modules (some modules might need an explicit import but are in the same namespace).

3

Oh dear. Could this be a feature that can be requested from Pytorch?

4

Sure, you can create this feature request on GitHub and depending on the interest, someone might want to work on it. However, note that I don’t see which subset should be built if you depend on import torch directly.

1 Like