4 ms·
I wonder if "faster" really matters. The implementation-level optimization seems paled when compared to productivity, ecosystem, and algorithm-level and hardwar
by hintymad 5y ago
I wonder if "faster" really matters. The implementation-level optimization seems paled when compared to productivity, ecosystem, and algorithm-level and hardware-level boost. Case in point, Caffe2 wanted to be the backend of PyTorch in FB, yet failed. MxNet claims to be "faster" than PyTorch in pretty much every way and Smola personally pushed orgs in AWS AI really hard to adopt MxNet and Gluon* yet failed. The latter failure is particularly worth mentioning: all teams in AWS AI complied and adopted MxNet but then three years later decided to move to PyTorch. All-in-all, the incremental speed boost was simply irrelevant.
- dr_zoidberg 5y agoI keep a few projects at work where we train models with Keras+TensorFlow. Twice before I tried to switch to PyTorch, because it did have nicer features/syntax for research and SOTA models, but deploy and support was limited on the toolchain we use, so we had to stay with Keras+TF. I'd still like to switch to PyTorch because of a few goodies, but the deploy of the models on some platforms is still simpler starting from a TF model. If anyone has a few links or resources to point me to, I'm more than happy to make a third try.
- stainablesteel 5y agoif there was constant deployment in some large industrial niche, faster implementations might help you save on your electric bill