added information about configuring PLUMED

plumed-libtorch-interface/README.md

@@ -1,5 +1,42 @@
 ## PLUMED - Pytorch interface
 
-This folder contains the C++ interface to load machine-learning models trained with Pytorch into PLUMED. 
+In order to use the Deep-TICA CVs to enhance the sampling, we need to load the NN models inside PLUMED.
 
-Please follow the previous tutorial on the [DeepLDA method](https://colab.research.google.com/github/luigibonati/data-driven-CVs/blob/master/code/Tutorial%20-%20DeepLDA%20training.ipynb#scrollTo=F7qSDdBGn8Vv) for instructions about how to configure PLUMED and install the PLUMED-torch interface. 
+This folder contains a Plumed-libtorch interface, as developed for the [Deep-LDA](https://github.com/luigibonati/data-driven-CVs) NN CVs.
+
+**(1) Download Libtorch**
+
+Prebuilt binaries can be downloaded from [Pytorch](https://pytorch.org/) website. Both binaries (cxx11 ABI and pre-cxx11) can be used. Note that the versions of Pytorch and LibTorch should match in order to load correctly the serialized model. The following instructions are related to [LibTorch 1.4](http://download.pytorch.org/libtorch/cpu/libtorch-shared-with-deps-1.4.0%2Bcpu.zip), which work with the model specified in the `mlcvs` requirements.
+
+**(2) Add PytorchModel.cpp to PLUMED**
+
+You need to add this the `PytorchModel.cpp` in the source directory of PLUMED2 (e.g. `plumed2/src/function/`) and, after configuring PLUMED with the Libtorch library, recompile PLUMED. 
+Alternatively, this interface can be also loaded in runtime using the LOAD command in the PLUMED input file: 
+```
+LOAD FILE=PytorchModel.cpp
+```
+Note that also in this second case you need to configure PLUMED to include the Libtorch library (see below), so I suggest to recompile PLUMED with the .cpp file in it, so that you can immediately detect if the linking was succesful.
+
+**(3) Configure PLUMED**
+
+If `$LIBTORCH` contains the location of the downloaded binaries, we can configure PLUMED in the following way:
+
+```
+./configure  --enable-rpath \
+             --disable-external-lapack --disable-external-blas \
+             CXXFLAGS="-O3 -D_GLIBCXX_USE_CXX11_ABI=0" \
+             CPPFLAGS="-I${LIBTORCH}/include/torch/csrc/api/include/ -I${LIBTORCH}/include/ -I${LIBTORCH}/include/torch" \
+             LDFLAGS="-L${LIBTORCH}/lib -ltorch -lc10 -Wl,-rpath,${LIBTORCH}/lib"
+```
+
+Notes:
+- This command is valid for the pre-cxx11 ABI version. If you downloaded the cxx11 ABI one the corresponding option should be enabled in the configure: `CXXFLAGS="-D_GLIBCXX_USE_CXX11_ABI=1"`
+- When using different versions of Libtorch the name of the libraries to be linked might be different.
+
+
+**(4) Load the model in the input file**
+
+In the PLUMED input file one should specify the model and the arguments. The interface detects the number of outputs and create a component for each of them, which can be accessed as cv.node-0, cv.node-1, ... 
+```
+cv: PYTORCH_MODEL MODEL=model.pt ARG=d1,d2,...,dN
+```