Update on "[ONNX] Do not run 'deduplicate_initializers' when 'keep_in…

…itializers_as_inputs' is True"


### Proposal
When arg of 'keep_initializers_as_inputs' is True, it's quite possible that parameters are set by initializer of input.
Hence we should disable de-duplicate initializer optimization when 'keep_initializers_as_inputs==True'.

- [ ] Update doc related to `keep_initializers_as_inputs`.

[ghstack-poisoned]

torch/onnx/utils.py

@@ -454,6 +454,11 @@ def forward(self, x):
             This may allow for better optimizations (e.g. constant folding) by
             backends/runtimes.
 
+            If True, `deduplicate_initializers` pass will not be executed. This means
+            initializers with duplicated values will not be deduplicated and
+            will be treated as distinct inputs to the graph. This allows different
+            input initializers to be supplied at the runtime following export.
+
             If ``opset_version < 9``, initializers MUST be part of graph
             inputs and this argument will be ignored and the behavior will be
             equivalent to setting this argument to True.