Fix shape mismatch caused by broadcast-to-TMS

[ashokkumarkannan1]

Ticket

Issue

Problem description

The convert_to_tms transformation during pre-lowering inserts a TMS operation into the cast node, which is unary. This leads to a verification issue, as the cast operation is expected to preserve the original shape after this change the following alvert and distilbert will be facing the config issue,and the post processing of those model is rectified in this #2858

What's changed

The Broadcast Op Mapping and attr mapping is added in lower_to_mlir and modified the Broadcast op in tm.py to include the broadcast_dimension attr and modified the pre_lowering_passes convert_to_tms function to include the check wether the consumer is unary op and then adding the tms by this changes the following models that is block will now execute end to end now.
Core Changes
MLIR Integration: Added Broadcast operation mapping and attribute mapping in lower_to_mlir
TM Operation Enhancement: Modified Broadcast op in tm.py to include broadcast_dimension attribute
Pre-lowering Logic: Enhanced convert_broadcast_ops_to_tms function with unary consumer detection and improved TM insertion
Frontend Node Detection: Added logic to distinguish between user-defined and frontend-transformed broadcast nodes using layer tag patterns

Model Log

bert_onnx.log
albert_basev1.log
clip_onnx.log
codegen_mono.log
codegen_multi.log
codegen_nl.log
distilbert_onnx.log
nanogpt_onnx.log

Checklist

• New/Existing tests provide coverage for changes

Example Transformation Comparison

Before Fix

After Fix

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 62.74%. Comparing base (856ebd6) to head (38eac39).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2827   +/-   ##
=======================================
  Coverage   62.74%   62.74%           
=======================================
  Files         156      156           
  Lines       13248    13248           
=======================================
  Hits         8312     8312           
  Misses       4936     4936           

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