Docs / jax4j-engine-cuda / com.nablatensor.jax4j.engine
final class
CudaEagerEngine
CUDA engine that launches one kernel per tape node, as an eager tensor framework would. Also compilation-free, and the slowest of the three CUDA engines by some margin; it is here as the baseline the other two are worth measuring against.
Methods
String name()
int priority()
boolean isAvailable()
boolean supports(AadOptions options)
String describe()
AadExecutable compile(AadTape tape, AadOptions options)