I presume this relates, directly or indirectly, to the Anaconda licensing changes.<p>Even though you can avoid the license issues (eg with conda-forge) it brought up a discussion where we looked into moving away and found we didn't face anything like the challenges we used to (which had originally brought us to conda) - feels like this may have backfired on Anaconda and perhaps that's borne out by the usage points the pytorch post mentions.