1

I am curious whether I can download big packages in batches using conda (like in web browsers where we can resume interrupted downloads).

I wanted to update my pytorch package but the size is too big for my limited mobile data.

The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    libuv-1.40.0               |       he774522_0         255 KB
    pytorch-1.8.1              |py3.7_cuda10.1_cudnn7_0 836.4 MB  pytorch
    torchvision-0.9.1          |       py37_cu101         7.3 MB  pytorch
    ------------------------------------------------------------
                                           Total:       843.9 MB

The following NEW packages will be INSTALLED:

  libuv              pkgs/main/win-64::libuv-1.40.0-he774522_0

The following packages will be UPDATED:

  pytorch                      1.4.0-py3.7_cuda101_cudnn7_0 --> 1.8.1-py3.7_cuda10.1_cudnn7_0
  torchvision                              0.5.0-py37_cu101 --> 0.9.1-py37_cu101
iacob
  • 14,010
  • 5
  • 54
  • 92
Srinjoy Mukherjee
  • 25
  • 1
  • 5
  • You could modify the script I wrote [in this answer](https://stackoverflow.com/a/66210871/570918) to download the tarballs using whatever tool you'd like. There, I used `wget` with the `-c` flag, which will resume partial downloads. The only additional step you'd need is to extract the downloaded packages to a package cache. – merv May 27 '21 at 17:02

1 Answers1

0

No this is not possible in conda. Completely downloaded files will not be redownloaded, but files which are only partially downloaded before interruption will need to be redownloaded:

https://github.com/conda/conda/issues/800

iacob
  • 14,010
  • 5
  • 54
  • 92
  • I’d qualify it with “*not possible with the Conda CLI*”. One could hack together a workaround to use `wget` with the continuation flag to do your downloading, building on what was demonstrated [in this answer](https://stackoverflow.com/a/66210871/570918). So, a custom FETCH procedure and let Conda still handle the solving and LINK procedures. – merv Jun 10 '21 at 20:29