[fx2trt] fix test failures (#65)

Summary:
Pull Request resolved: pytorch/fx2trt#65

pytorch/pytorch#73433 changes the default argument value of `value` in functional.pad from `0` to `None` which breaks fx2trt tests.

Reviewed By: frank-wei, wushirong

Differential Revision: D36076942

fbshipit-source-id: f6f6f869deb106e90b484fc07574790c0b6df46e

Author: 842974287

fx/converters/acc_ops_converters.py

@@ -261,7 +261,7 @@ def acc_ops_pad_with_padding_layer(
     input_val = kwargs["input"]
     pad = cast(Sequence[int], kwargs["pad"])
     mode = kwargs["mode"]
-    value = kwargs["value"]
+    value = kwargs["value"] if kwargs["value"] is not None else 0
     rank = len(input_val.shape)  # type: ignore[union-attr]
 
     if not isinstance(input_val, TRTTensor):
@@ -311,7 +311,7 @@ def acc_ops_pad_with_slice_layer(
     input_val = kwargs["input"]
     pad = cast(Sequence[int], kwargs["pad"])
     mode = kwargs["mode"]
-    value = kwargs["value"]
+    value = kwargs["value"] if kwargs["value"] is not None else 0
     rank = len(input_val.shape)  # type: ignore[union-attr]
 
     if not isinstance(input_val, TRTTensor):