Jacob Kahn

To follow up here, we're currently working on supporting OpenCL training on the CPU, which ArrayFire does in a parallelized fashion. This should dramatically speed up CPU training and inference....

Integrated elsewhere.

@loretoparisi @lengoanhcat -- we've recently spun off Shumai (https://github.com/facebookresearch/shumai) which binds Flashlight to the Bun Javascript; we're in the process of building NodeJS bindings as well.

Done via Flashlight Text.

@HalFTeen — you need gcc >= 7 realistically to compile the project. When you set the path, make sure you've removed the `build` directory or wherever you performed your CMake...

@LBY-bamboo -- the error message describes the issue. You need to select which ArrayFire backend will be used since multiple viable options were found.

@LBY-bamboo -- that CMake option is deprecated. You can use the following options: ```cmake -DFL_USE_ARRAYFIRE=ON -DFL_ARRAYFIRE_USE_CUDA=ON ``` to use the ArrayFire CUDA backend. Swap to `-DFL_USE_ARRAYFIRE_CPU=ON` to use the ArrayFire...

@bioinfornatics — sorry for the delay; this is a work in progress. This will require adding new tensor backends that support RoCM backends. There are a few routes to this:...

@yurivict oneDNN breaks their API with new releases -- I'm working on supporting 3.0+. Thanks for reporting.

@tlyong3 — this continues to be a work in progress. We're awaiting some more correctness checks on our end and ensuring oneDNN 3.0 works on all of our build environments;...