IRC channel logs


back to list of logs

<rekado>I’d like us to make some progress towards tensorflow 2 again
<rekado>which means that we need bazel
<rekado>can we build it first by adding all its maven inputs to the native inputs first?
<rekado>I just want to build it *first* and substitute jars later
<rekado>I gotta say: this Python world of huge blobs of “open source” software is wearing me down
<rekado>just import InstructorEmbedding you say?
<rekado>so easy!
<rekado>well, it needs pytorch and datasets and sentence-transformers, and these need huggingface-hub, and tensorflow (because a few more gigs in dependencies are barely noticeable when we already have pytorch)
<rekado>and apache spark, and also tensorflow 2
<rekado>and spacy, but also spacy_legacy for good measure…
<rekado>and some other thing that embeds chromium for rendering SVG
<rekado>and here’s blis but instead of just binding to blis we just bundle blis and build it all again
<rekado>I’m not complaining that there’s so much free software out there that happens to be written in Python
<rekado>but I seem to be incompatible with this maximalist approach of bundling everything and pulling in huge libraries, because “we can just build a statically linked blob and put this all in a wheel”
<rekado>I feel like I can’t do this any longer, because I’m out of touch.
<rekado>(it’s weird that I as an Emacs user have a problem with the kitchen sink approach to software development, but here we are)