You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
the cast operator only supports casting to fp16, fp32, i32, bool.
None of the operations support int64 as input(so the output of argMax/argMin can't be connected to the next layer if it's int64)
Graph output doesn't support i64. Only fp16, fp32, i32 is supported.
Given int32 is the intersection across these backends, I see two options to make it work:
a. Update argMin/Max to always output int32
b. On the spec level, allow passing a param of output_type. And allow probing the supported data types for this param.
The text was updated successfully, but these errors were encountered:
philloooo
changed the title
Consider changing output type of ArgMax/Argmin to int32, or allow passing output_type for ArgMax/ArgMin
Consider changing output type of ArgMax/Argmin to int32, or allow passing output_type
Apr 23, 2024
Following up from the working group meeting, I am happy to explore option 2 if @fdwr can provide examples on when int64 is useful.
I didn't fully understand your gather example, because I actually don't understand why indices allow both int64 and uint32. The indices should point to valid indices that's within MLOperand's dimensions right? And the dimensions are uint32...
Currently WebNN specifies ArgMax/Min returns int64.
Returning int64 can't be emulated on CoreML:
Given int32 is the intersection across these backends, I see two options to make it work:
a. Update argMin/Max to always output int32
b. On the spec level, allow passing a param of
output_type
. And allow probing the supported data types for this param.The text was updated successfully, but these errors were encountered: