Is there a tutorial / sample repo somewhere that solves this problem:
- you have a machine with some nvidia GPUs
- you have CUDA installed
- you follow this tutorial / install this package
- now you can use OCaml.xyz / OWL with GPGPU support ?
The closest to this seems to be; write code in OWL, export a ONNX, run some python/julia script to read/optimize ONNX, load weights back into OWL. This seems to limit us to situations where the compute graph is static …