Disable CW in manual script due to profiling issue (#1818)

Summary:
Pull Request resolved: #1818

Disable Columnwise Sharding in manual script as profiling takes a massive amount of time (result is a huge profile). This will not affect the automatic benchmark

Differential Revision: D55173220

fbshipit-source-id: c1bbfafeeda49ceb2a8ab224931247b357ea8fd6

torchrec/distributed/benchmark/benchmark_inference.py

@@ -44,7 +44,8 @@
 BENCH_SHARDING_TYPES: List[ShardingType] = [
     ShardingType.TABLE_WISE,
     ShardingType.ROW_WISE,
-    ShardingType.COLUMN_WISE,
+    # ShardingType.COLUMN_WISE,
+    # TODO: CW with FXJIT takes long time while profiling, doesn't cause an issue with no profiling in automatic benchmark
 ]
 
 BENCH_COMPILE_MODES: List[CompileMode] = [