[fx2trt] Fix dummy weight initialization in conv1d converter (#78402)

Summary:
X-link: pytorch/pytorch#78402

Pull Request resolved: pytorch/fx2trt#84

att, currently it errors out with the following error:
```
---> 72         dummy_weight = trt.Weights(weight_shape)
     73         layer = network.add_convolution_nd(
     74             input=input_val,
TypeError: __init__(): incompatible constructor arguments. The following argument types are supported:
    1. tensorrt.tensorrt.Weights(type: tensorrt.tensorrt.DataType = <DataType.FLOAT: 0>)
    2. tensorrt.tensorrt.Weights(a: numpy.ndarray)
```
full error: https://www.internalfb.com/phabricator/paste/view/P503598381
we need to pass arond a numpy ndarray instead of a shape here.

Reviewed By: wushirong

Differential Revision: D36721313

fbshipit-source-id: c13583c3accb37f781429b5905cfda8e856cebbc

Author: jerryzh168

fx/converters/acc_ops_converters.py

@@ -65,11 +65,11 @@ def acc_ops_conv1d(
         unsqueeze_weight_layer = network.add_shuffle(input=weight)
         unsqueeze_weight_layer.reshape_dims = tuple([*weight.shape, 1])
         set_layer_name(unsqueeze_layer, target, name + "_unsqueeze_weight")
-        weiht = unsqueeze_weight_layer.get_output(0)
+        weight = unsqueeze_weight_layer.get_output(0)
         weight_shape = tuple(kwargs["weight"].shape)  # type: ignore[union-attr]
         # will need to use uninitialized weight and set it later to support
         # ITensor weights
-        dummy_weight = trt.Weights(weight_shape)
+        dummy_weight = trt.Weights()
         layer = network.add_convolution_nd(
             input=input_val,
             num_output_maps=weight.shape[0],

test/quant/test_quant_trt.py

@@ -473,7 +473,7 @@ def forward(self, x):
 
         # just testing conv2d since conv1d and conv3d are not supported in fx2trt
         for dim, has_relu, f_relu, is_qat in itertools.product(
-            [2], [True, False], [True, False], [True, False]
+            [1, 2], [True, False], [True, False], [True, False]
         ):
             # when has_relu=False, we have torch.nn.Identity, which would introduce
             # extra quant-dequat pair
@@ -564,7 +564,6 @@ def forward(self, x):
             ns.call_module(torch.nn.quantized._reference.Linear): 1,
             ns.call_module(torch.nn.quantized._reference.Conv2d): 1,
         }
-        print(m)
         self.checkGraphModuleNodes(m, expected_node_occurrence=expected_occurrence)
 
     def test_unsupported_qconfig(self):
@@ -725,7 +724,6 @@ def conv_add_extra_inputs_getter(pattern):
             example_inputs,
             backend_config_dict=modified_backend_config_dict,
         )
-        print(m)
         node_occurrence = {
             ns.call_module(torch.ao.quantization.HistogramObserver): 3,
         }