No thanks to the wonderful tensorflow profiler documentation displaying features rather than how the actual thing can be used. I'm experiencing a memory leak which I described earlier in my post. I've been trying to find a way to profile memory of the things going on inside a tf.function ever since. But as it has always been the case, the documentation is as useless as never existing. For those who ever experienced memory leaks while using tf.function, is there a way to actually see what memory is allocated to which objects? Forget about tracemalloc, objgraph, memory-profiler which won't work, and of course the official tensorflow profiler which doesn't have a documentation. Let's say the leak won't happen unless there is a tf.function somewhere, what are the options here?
Asked
Active
Viewed 240 times
4
watch-this
- 1
- 4
- 20