feat: Add example usage scripts for dynamo path

- Add sample scripts covering resnet18, transformers, and custom
examples showcasing the `torch_tensorrt.dynamo.torch_compile` path,
which can compile models with data-dependent control flow and other such
restrictions which can make other compilation methods more difficult
- Cover different customizeable features allowed in the new backend

Author: gs-olive

examples/dynamo/torch_compile_advanced_usage.py

@@ -0,0 +1,53 @@
+import torch
+from torch_tensorrt.dynamo.torch_compile import create_backend
+from torch_tensorrt.fx.lower_setting import LowerPrecision
+
+
+##### Overview
+# This script is intended as an overview of the process by which
+# torch_tensorrt.dynamo.torch_compile works, and how it integrates
+# with the new torch.compile API.
+
+# We begin by defining a model
+class Model(torch.nn.Module):
+    def __init__(self) -> None:
+        super().__init__()
+        self.relu = torch.nn.ReLU()
+
+    def forward(self, x: torch.Tensor, y: torch.Tensor):
+        x_out = self.relu(x)
+        y_out = self.relu(y)
+        x_y_out = x_out + y_out
+        return torch.mean(x_y_out)
+
+
+##### Compilation using default settings
+
+sample_inputs = [torch.rand((5, 7)).cuda(), torch.rand((5, 7)).cuda()]
+model = Model().eval().cuda()
+
+# Next, we compile the model using torch.compile
+# For the default settings, we can simply call torch.compile
+# with the backend "tensorrt", and run the model on an
+# input to cause compilation, as so:
+optimized_model = torch.compile(model, backend="tensorrt")
+optimized_model(*sample_inputs)
+
+
+##### Compilation using custom settings
+
+sample_inputs_half = [
+    torch.rand((5, 7)).half().cuda(),
+    torch.rand((5, 7)).half().cuda(),
+]
+model_half = Model().half().eval().cuda()
+
+# Alternatively, if we want to customize certain options in the backend,
+# but still use the torch.compile call directly, we can call the
+# convenience/helper function create_backend to create a custom backend
+# which has been pre-populated with certain key
+custom_backend = create_backend(
+    lower_precision=LowerPrecision.FP16, debug=True, max_num_trt_engines=2
+)
+optimized_model_custom = torch.compile(model_half, backend=custom_backend)
+optimized_model_custom(*sample_inputs_half)

examples/dynamo/torch_compile_resnet_example.py

@@ -0,0 +1,45 @@
+import torch
+from torch_tensorrt.dynamo import torch_compile
+import torchvision.models as models
+
+##### Overview
+# This script is intended as a sample of the torch_tensorrt.dynamo.torch_compile
+# workflow on the resnet18 model
+
+
+# Initialize model and sample inputs
+model = models.resnet18(pretrained=True).half().eval().to("cuda")
+inputs = [torch.randn((1, 3, 224, 224)).to("cuda").half()]
+
+##### Optional Input Arguments
+
+# Enabled precision for TensorRT optimization
+enabled_precisions = {torch.half}
+# Whether to print verbose logs
+debug = True
+# Workspace size for TensorRT
+workspace_size = 20 << 30
+# Maximum number of TRT Engines
+# (Higher value allows more graph segmentation)
+max_num_trt_engines = 100
+
+
+# Build and compile the model with torch.compile, using tensorrt backend
+optimized_model = torch_compile(
+    model,
+    inputs,
+    enabled_precisions=enabled_precisions,
+    debug=debug,
+    workspace_size=workspace_size,
+    max_num_trt_engines=max_num_trt_engines,
+)
+
+
+# Does not cause recompilation (same batch size as input)
+new_inputs = [torch.randn((1, 3, 224, 224)).half().to("cuda")]
+new_outputs = optimized_model(*new_inputs)
+
+
+# Does cause recompilation (new batch size)
+new_batch_size_inputs = [torch.randn((8, 3, 224, 224)).half().to("cuda")]
+new_batch_size_outputs = optimized_model(*new_batch_size_inputs)

examples/dynamo/torch_compile_transformers_example.py

@@ -0,0 +1,49 @@
+import torch
+from torch_tensorrt.dynamo import torch_compile
+from transformers import BertModel
+
+##### Overview
+# This script is intended as a sample of the torch_tensorrt.dynamo.torch_compile
+# workflow on the BERT base uncased model
+
+
+model = BertModel.from_pretrained("bert-base-uncased").cuda().eval()
+inputs = [
+    torch.randint(0, 1, (1, 14), dtype=torch.int32).to("cuda"),
+    torch.randint(0, 1, (1, 14), dtype=torch.int32).to("cuda"),
+]
+
+##### Optional Input Arguments
+
+# Enabled precision for TensorRT optimization
+enabled_precisions = {torch.float}
+# Whether to print verbose logs
+debug = True
+# Workspace size for TensorRT
+workspace_size = 20 << 30
+# Maximum number of TRT Engines
+# (Higher value allows more graph segmentation)
+max_num_trt_engines = 200
+
+
+# Build and compile the model with torch.compile, using tensorrt backend
+optimized_model = torch_compile(
+    model,
+    inputs,
+    enabled_precisions=enabled_precisions,
+    debug=debug,
+    workspace_size=workspace_size,
+    max_num_trt_engines=max_num_trt_engines,
+)
+
+# Does not cause recompilation (same batch size as input)
+new_inputs = [
+    torch.randint(0, 1, (1, 14), dtype=torch.int32).to("cuda"),
+    torch.randint(0, 1, (1, 14), dtype=torch.int32).to("cuda"),
+]
+new_outputs = optimized_model(*new_inputs)
+
+
+# Does cause recompilation (new batch size)
+new_batch_size_inputs = [torch.randn((8, 3, 224, 224)).half().to("cuda")]
+new_batch_size_outputs = optimized_model(*new_batch_size_inputs)