[torchlib] Fix aten::arange to support dynamic shapes

[titaiwangms]

Runinng SmolLM_1_7b with dynamic shapes discovered that arange.start was not dynamic anymore, since https://github.com/microsoft/onnxscript/pull/1781/files.

[justinchuby]

Thanks, also interesting - does this mean we need to update https://github.com/pytorch/pytorch/blob/10f16cc7daf89a73c1177dfbd0bb0fbb0fc8683b/aten/src/ATen/native/native_functions.yaml#L780 ? Should we create an issue for propose a fix?

[justinchuby]

Should we also fix the rest of the arange overloads?

[codecov]

Codecov Report

Attention: Patch coverage is 66.66667% with 12 lines in your changes missing coverage. Please review.

Project coverage is 75.45%. Comparing base (f18dadc) to head (0c9c48e).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
onnxscript/function_libs/torch_lib/ops/core.py	66.66%	8 Missing and 4 partials ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1913   +/-   ##
=======================================
  Coverage   75.45%   75.45%           
=======================================
  Files         252      252           
  Lines       27327    27335    +8     
  Branches     3172     3171    -1     
=======================================
+ Hits        20619    20626    +7     
  Misses       5745     5745           
- Partials      963      964    +1     

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

[titaiwangms]

Should we also fix the rest of the arange overloads?

Sure, I can update the whole family. And record the question that does Scalar cover sym_int and sym_float in Meta doc.

[xadupre]

Is it possible to add a unit test to make sure there is no regression about this?

[titaiwangms]

Is it possible to add a unit test to make sure there is no regression about this?

I will merge this first to unblock the model, and then come up with tests as follow-up.

[justinchuby]

The old arange logic had issues with inconsistent types. Please make sure that's resolved. I am happy if you fix one and take time to ensure the rest of the variants are correct in a separate PR, if that allows faster progress