packaging: Specify cpuonly for conda meta.yaml

[seemethere]

Package installation for pytorch binaries was failing out due to
conda/conda-package-handling#71

Default to cpuonly since that won't fail out.

Similar to pytorch/audio#1105

This should resolve issues within the nightly pipeline for conda builds

Signed-off-by: Eli Uriegas [email protected]

[xuzhao9]

Sorry I am trying to understand what does "cpuonly" mean in this PR? Does it mean we only upload cpu version of torchtext nightly to conda (and by default conda install torchtext -c pytorch-nightly won't have gpu support)?

[seemethere]

Sorry I am trying to understand what does "cpuonly" mean in this PR? Does it mean we only upload cpu version of torchtext nightly to conda (and by default conda install torchtext -c pytorch-nightly won't have gpu support)?

cpuonly is a metapackage that we add as a dependency for cpu versions of pytorch

Basically the way the conda dependency resolver works is that if we specify that we explicitly want cpuonly installed as a package then it should pick the pytorch package that has cpuonly as a dependency (which is our cpu specific package).

The same works for cudatoolkit versions as well, so for example if we have conda install -y -c pytorch pytorch cudatoolkit=11.0 it should install the pytorch package that is compiled with CUDA 11.0 support.

[seemethere]

Actually, @xuzhao9 made me realize that this might not be the best solution since it'd limit the pytorch version you can install with torchtext to one that depends on cpuonly, meaning it'd only install the cpu specific version, let me see if I can resolve this

[seemethere]

Okay I made a change that puts in more in line with what torchaudio is doing which shouldn't limit what people can install with torchtext

[codecov]

Codecov Report

Merging #1106 (f1c7f7d) into master (09f1b85) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master    #1106   +/-   ##
=======================================
  Coverage   77.54%   77.54%           
=======================================
  Files          45       45           
  Lines        3086     3086           
=======================================
  Hits         2393     2393           
  Misses        693      693           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 09f1b85...f1c7f7d. Read the comment docs.

[xuzhao9]

Looks good!