There's a new proof. Does not seem to require too much prior knowledge. Based on plays with boolean function normal forms (conjunctive and disjunctive).
After two days of struggling, I was finally able to build libtensorflow.dll (v1.2.1) for Windows with GPU (CUDA 8 + CUDNN 6) support. TL;DR; here's the file (~160MB). It seems (so far) usable with TensorFlowSharp from Nuget, but you have to manually swap DLL in the package cache. UPD 2019-03-29 : instead of using TensorFlowSharp , I am now using Gradient - it provides access to the full Python API. And you don't have to manually build TensorFlow for GPU - just install Python 3.6, and follow the official TensorFlow instructions to install tensorflow 1.10 or tensorflow-gpu 1.10, or tensorflow-rocm for ATI. Gradient picks it up automatically or via GradientSetup class . Some notes on the build (in case you want to reproduce it): I used Visual Studio 2017 Despite mentioning only VS 2015 C++ compiler as compatible, I was able to build with VS 2017 compiler. Build with CMake CMake files are in ./tensorflow/contrib/cmake Do not forget SWIG Download and install SWIG.
The amount of committed memory (as seen in Task Manager -> Performance -> Memory) would slowly grow to unrealistically high values. One time, after it crossed 100GB (I have just 32GB of RAM), it pushed my pagefile.sys into 100GB+ too, which consumed almost all remaining space on the SSD. Neither Task Manager, nor even SysInternals Process Explorer showed anything suspicious, but I still decided to walk through and shutdown unused services. Turns out it was RunSW Windows service by Realtek, that has something to do with USB WiFi dongles (I stopped using one over 4 months ago). I was not able to find how to completely remove the service, but stopping, and disabling it instantly solved the problem (committed dropped to ~20GB), and the system seems to work fine now.
Комментарии
Отправить комментарий