Joe Wallwork
Joe Wallwork
``` -- Building with Fortran PyTorch coupling -- Configuring done (20.6s) -- Generating done (0.2s) -- Build files have been written to: D:/a/FTorch/FTorch/build [ 5%] Building Fortran object CMakeFiles/ftorch.dir/src/ftorch.F90.obj [...
e.g., It'd be useful to be able to get the available devices and also to get the device type associated with a given tensor.
Towards #111. We should further test #152 by calling `tensor.backward` on a tensor outputted from a network. By doing so, we'll check that we're able to back-propagate through a network...
There was an unfinished C and C++ example in FTorch for some time. The draft code can be found in #230, which removed this example. Not for this PR but...
One observation (that is not to be resolved here) is that our [online API docs](https://cambridge-iccs.github.io/FTorch/lists/procedures.html) are no longer very useful for basic users and are instead geared at developers. In...
Do you think it might be worth bumping the minimum CMake version to 3.18, as they have in [PyTorch](https://github.com/pytorch/pytorch/blob/main/torch/CMakeLists.txt)? I think 3.15 still works in most cases, but [CMake 3.18...
This will be useful for computing cost functions on the FTorch side.
I had a brainwave: if we overload each binary elementary operation (+, -, *, /) so that it can accept a tensor and a Fortran array then we can apply...
I imagine for realistic cases, we'll want to be able to take slices of output data. As such, we'll need to define what is meant by slicing of a tensor.