Skip inserting observers for Tensors inside fused op (#25281)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/25281
We want to skip inserting observers for the Tensors that's between the two
ops that will be fused, e.g. Conv -> ReLU, this PR just added this pattern,
but new patterns can be easily added in the future.
Test Plan:
python test test/test_jit.py -- 'TestJit.test_insert_observers_skip_values'
Imported from OSS
Differential Revision: D17106037
fbshipit-source-id: 49697f4d9598a461edc62a2b4148495764a99574
