comaniac commented on pull request #8775:
URL: https://github.com/apache/tvm/pull/8775#issuecomment-902096517
In TE compiler, the input tensors are internally created by
`ScheduleBuilder`. Thus, it could be a common case for developers who want to
lower a Relay function to TE but don't have tensor inputs (i.e.,
`te.placeholder` instead of `relay.var`). IMHO, one exciting feature of compile
engine refactoring is to decouple TE lowering and scheduling, so that
developers can get the lowered TE before scheduling, so it would be great if we
could at least expose the interface like the following to developers:
```
te_func = lower(relay_func, target)
tensors = te_func.outputs
for tensor in tensors:
print(tensor) # Print the TE compute AST before scheduling.
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
