Closed philloooo closed 4 months ago
It occurs to me the existing verbiage "Implementations may impose a smaller upper bound" technically already covered this case, since CoreML and TFLite impose a smaller bound (2 billion) than the DML EP (4 billion). Though, I'm content with this restriction, since 2 billion is well more than enough for a single dimension (never saw a model that even came close to that in half a decade for a single tensor).
fixes #734
Update
valid dimension
definition to be in the range oflong
and greater than 0.@fdwr @huningxin
Preview | Diff