Conversation
There was a problem hiding this comment.
Pull Request Overview
This PR introduces a new user guide for comparing TraceLens performance reports and refactors tensor-type detection logic in the performance model.
- Add
examples/compare_perf_reports.mdwith step-by-step instructions and tables for comparing.xlsxreports. - Remove the
is_tensortypehelper inTraceLens/PerfModel/perf_model.pyand replace it with shape-based checks.
Reviewed Changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| examples/compare_perf_reports.md | New markdown guide detailing inputs, flags, output, and design. |
| TraceLens/PerfModel/perf_model.py | Remove is_tensortype; use shape_in* != () for tensor checks. |
| in1_is_tensor = self.param_details['shape_in1'] != () | ||
| in2_is_tensor = self.param_details['shape_in2'] != () |
There was a problem hiding this comment.
The original is_tensortype check also verified the data type; switching to shape-only may misclassify non-floating tensors (e.g., integer tensors) as 'tensor types'. Consider reintroducing a dtype check or renaming this logic to accurately reflect its new semantics.
| in1_is_tensor = self.param_details['shape_in1'] != () | |
| in2_is_tensor = self.param_details['shape_in2'] != () | |
| in1_is_tensor = self.param_details['shape_in1'] != () and name2bpe(self.param_details['dtype_in1_in2_out'][0]) is not None | |
| in2_is_tensor = self.param_details['shape_in2'] != () and name2bpe(self.param_details['dtype_in1_in2_out'][1]) is not None |
|
|
||
| --- | ||
|
|
||
| ### 1 . What It Takes In |
There was a problem hiding this comment.
[nitpick] There is an extra space before the period in the section header. For consistency, change ### 1 . What It Takes In to ### 1. What It Takes In.
| ### 1 . What It Takes In | |
| ### 1. What It Takes In |
No description provided.