3 ms·
So this exploits the fact that floating point numbers have finite precision (and perhaps uneven spacing) to generate non-linear operations? That's actually a r
by SomeStupidPoint 9y ago
So this exploits the fact that floating point numbers have finite precision (and perhaps uneven spacing) to generate non-linear operations?
That's actually a really cool usage of the specification!
- dbcurtis 9y agoWell, except that the author misunderstands the specification and how it is typically implemented on modern computers.
- gdb 9y ago(I work at OpenAI.) TensorFlow by default is built with denormals off (ftz=true), so denormals aren't relevant for the applications we're interested in. We have updated the post to indicate this — thanks for the feedback!