Add aten_maximum_real

[xadupre]

No description provided.

[codecov]

Codecov Report

Attention: Patch coverage is 66.66667% with 1 line in your changes missing coverage. Please review.

Project coverage is 70.14%. Comparing base (51ecf47) to head (2c7908a).
Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
onnxscript/function_libs/torch_lib/ops/core.py	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2371      +/-   ##
==========================================
- Coverage   70.14%   70.14%   -0.01%     
==========================================
  Files         197      197              
  Lines       24983    24986       +3     
  Branches     2667     2667              
==========================================
+ Hits        17525    17526       +1     
- Misses       6531     6533       +2     
  Partials      927      927              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[justinchuby]

The types should be promoted already when the op is being translated and should be handled by

onnxscript/onnxscript/function_libs/torch_lib/ops/core.py

Line 5319 in 51ecf47

def aten_maximum(self: TReal, other: TReal) -> TReal:

. Was there a case when this isn't true? ONNX max requires all inputs to be the same type https://onnx.ai/onnx/operators/onnx__Max.html but this change allows the two inputs to take different types, which would allow an invalid model

[justinchuby]

Explained above and for further discussions

[xadupre]

There is a regression for the model XGLMForCausalLM. I tried this change based on the error message I got (missing signature for aten::max.other). But type promotion fails as you mentioned. Error: Type parameter (T) of Optype (Max) bound to different types (tensor(float16) and tensor(float) in node (n0).

[justinchuby]

Looks like max.other is an alias of maximum (https://github.com/pytorch/pytorch/blob/b44306d3681d5b248e6b439d293ea0d5a8903a61/aten/src/ATen/native/native_functions.yaml#L10064-L10069). I think we can do one of two things:

1. Unregister max.other and hope that it gets decomposed to maximum, which we have a type promotion rule for.
2. Add max.other to entries in https://github.com/pytorch/pytorch/blob/b44306d3681d5b248e6b439d293ea0d5a8903a61/torch/onnx/_internal/fx/passes/type_promotion.py#L828-L833

[titaiwangms]

Looks like max.other is an alias of maximum (https://github.com/pytorch/pytorch/blob/b44306d3681d5b248e6b439d293ea0d5a8903a61/aten/src/ATen/native/native_functions.yaml#L10064-L10069). I think we can do one of two things:

1. Unregister max.other and hope that it gets decomposed to maximum, which we have a type promotion rule for.
2. Add max.other to entries in https://github.com/pytorch/pytorch/blob/b44306d3681d5b248e6b439d293ea0d5a8903a61/torch/onnx/_internal/fx/passes/type_promotion.py#L828-L833

I will take a look and update it here.

[titaiwangms]

#2377 should fix what this PR tried to fix.