[Prim] Filter tensor type for int_array and scalar input in composite rule #51208
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
Actually, tensor type for int_arrary and scalar is not legacy type, it just not suitable for composite for now. We may find a way to limit this during dy2st process |
| raise ValueError( | ||
| f'No value found for {arg_name} of {arg_type} type for operator {op.type}.' | ||
| ) | ||
| for item in ("tensor_name", "tensor_name"): |
There was a problem hiding this comment.
it's a typo. Fixed.
cyber-pioneer
force-pushed
the
int_array
branch
from
c7b1108 to
848f6e7
Compare
cyber-pioneer
force-pushed
the
int_array
branch
from
848f6e7 to
2a78275
Compare
PR types
Bug fixes
PR changes
Others
Describe
Tensor type for int_array and scalar input will cause dynamic shape in static mode, but composite rule does not support currently.
As for arg "shape" of int_array type. It may be one of [list/tuple of int, tensor]. The former is widely used actually, while the latter is rarely used and will cause dynamic shape in static mode which needs extra unnecessary process. Thus, it would be better to filter it in composite rules.