Investigate Failures in Torch Benchmarks caused by #3104

[shivadbhavsar]

3104 is a big change and exposes some bugs thats causing several torch benchmark models to fail. Create smaller repros for these failures.

• [ ] has: convolution: Wrong number of arguments: expected 2 but given 1
• [ ] CONCAT: all input dimensions should match along axis 0

[shivadbhavsar]

Tracked in individual issues: #3218, #3224