4 ms·
Good question. I wonder if the point of this article is that other tools can be used to access TPUs other than the nominal Tensorflow-based workflow. I.e. the
by chrispeel 8y ago
Good question.
I wonder if the point of this article is that other tools can be used to access TPUs other than the nominal Tensorflow-based workflow. I.e. the article is to show that another workflow exists to access the capabilities of the TPU, which are (to some degree at least) already known. Keno Fischer says [1] they'll have other benchmarks out soon.
[1] https://twitter.com/KenoFischer/status/1054932277824774145 https://twitter.com/KenoFischer/status/1054932277824774145
- kwertzzz 8y agoOK, I see. In any case, even if the performance would not match Tensforflow on TPU this is a major achievement. After all, Google made the TPUs specifically for Tensorflow. I would not have imagined that it was so "easy" to replace LLVM by XLA.