fix: Allow low rank inputs in Python Runtime (#2282)

Author: gs-olive

py/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.py

@@ -1,13 +1,14 @@
 from __future__ import annotations
 
+import logging
 from typing import Any, Dict, List, Optional, Sequence, Tuple
 
+import tensorrt as trt
 import torch
 from torch.nn import Module
 from torch_tensorrt.fx.utils import Frameworks, unified_dtype_converter
 
-# @manual=//deeplearning/trt/python:py_tensorrt
-import tensorrt as trt
+logger = logging.getLogger(__name__)
 
 
 class PythonTorchTensorRTModule(Module):  # type: ignore[misc]
@@ -22,14 +23,12 @@ def __init__(
         engine: trt.ICudaEngine,
         input_names: Optional[List[str]] = None,
         output_names: Optional[List[str]] = None,
-        cuda_graph_batch_size: int = -1,
     ):
         super(PythonTorchTensorRTModule, self).__init__()
         self._register_state_dict_hook(PythonTorchTensorRTModule._on_state_dict)
         self.engine = engine
         self.input_names = input_names if input_names is not None else []
         self.output_names = output_names if output_names is not None else []
-        self.cuda_graph_batch_size = cuda_graph_batch_size
         self.initialized = False
         self._initialize()
 
@@ -107,7 +106,6 @@ def _on_state_dict(self, state_dict: Dict[str, Any], prefix: str, _: Any) -> Non
         state_dict[prefix + "engine"] = bytearray(self.engine.serialize())
         state_dict[prefix + "input_names"] = self.input_names
         state_dict[prefix + "output_names"] = self.output_names
-        state_dict[prefix + "cuda_graph_batch_size"] = self.cuda_graph_batch_size
 
     def _load_from_state_dict(
         self,
@@ -156,8 +154,6 @@ def forward(self, *inputs: Any) -> torch.Tensor | Tuple[torch.Tensor, ...]:
                     self.input_names
                 ), f"Wrong number of inputs, expect {len(self.input_names)} get {len(inputs)}."
 
-                # This is only used when the trt engine is using implicit batch dim.
-                batch_size = inputs[0].shape[0]
                 contiguous_inputs: List[torch.Tensor] = [i.contiguous() for i in inputs]
                 bindings: List[Any] = [None] * (
                     len(self.input_names)
@@ -166,25 +162,29 @@ def forward(self, *inputs: Any) -> torch.Tensor | Tuple[torch.Tensor, ...]:
                 )
 
                 for i, input_name in enumerate(self.input_names):
-                    assert inputs[
-                        i
-                    ].is_cuda, f"{i}th input({input_name}) is not on cuda device."
+                    if not contiguous_inputs[i].is_cuda:
+                        logger.warning(
+                            f"Detected input {input_name} of engine {self.engine.name} is not on a cuda device. "
+                            "This tensor is being moved by the runtime but for performance considerations, "
+                            "ensure your inputs are all on GPU and open an issue here "
+                            "(https://github.com/pytorch/TensorRT/issues) if this warning persists."
+                        )
+                        contiguous_inputs = (
+                            contiguous_inputs[:i]
+                            + [contiguous_inputs[i].cuda()]
+                            + contiguous_inputs[i + 1 :]
+                        )
+
                     assert (
-                        inputs[i].dtype == self.input_dtypes[i]
-                    ), f"Dtype mismatch for {i}th input({input_name}). Expect {self.input_dtypes[i]}, got {inputs[i].dtype}."
+                        contiguous_inputs[i].dtype == self.input_dtypes[i]
+                    ), f"Dtype mismatch for {i}th input({input_name}). Expect {self.input_dtypes[i]}, got {contiguous_inputs[i].dtype}."
 
                     idx = self.input_binding_indices_in_order[i]
                     bindings[idx] = contiguous_inputs[i].data_ptr()
 
-                    if not self.engine.has_implicit_batch_dimension:
-                        self.context.set_binding_shape(
-                            idx, tuple(contiguous_inputs[i].shape)
-                        )
-                    else:
-                        assert inputs[i].size()[1:] == self.input_shapes[i], (
-                            f"Shape mismatch for {i}th input({input_name}). "
-                            f"Expect {self.input_shapes[i]}, got {inputs[i].size()[1:]}."
-                        )
+                    self.context.set_binding_shape(
+                        idx, tuple(contiguous_inputs[i].shape)
+                    )
 
             with torch.autograd.profiler.record_function(
                 "PythonTorchTensorRTModule:ProcessOutputs"
@@ -193,10 +193,7 @@ def forward(self, *inputs: Any) -> torch.Tensor | Tuple[torch.Tensor, ...]:
                 outputs: List[torch.Tensor] = []
 
                 for i, idx in enumerate(self.output_binding_indices_in_order):
-                    if self.engine.has_implicit_batch_dimension:
-                        shape = (batch_size,) + self.output_shapes[i]
-                    else:
-                        shape = tuple(self.context.get_binding_shape(idx))
+                    shape = tuple(self.context.get_binding_shape(idx))
 
                     output = torch.empty(
                         size=shape,
@@ -207,10 +204,7 @@ def forward(self, *inputs: Any) -> torch.Tensor | Tuple[torch.Tensor, ...]:
                     bindings[idx] = output.data_ptr()
 
                 for i, idx in enumerate(self.hidden_output_binding_indices_in_order):
-                    if self.engine.has_implicit_batch_dimension:
-                        shape = (batch_size,) + self.hidden_output_shapes[i]
-                    else:
-                        shape = tuple(self.context.get_binding_shape(idx))
+                    shape = tuple(self.context.get_binding_shape(idx))
 
                     output = torch.empty(
                         size=shape,
@@ -222,14 +216,9 @@ def forward(self, *inputs: Any) -> torch.Tensor | Tuple[torch.Tensor, ...]:
             with torch.autograd.profiler.record_function(
                 "PythonTorchTensorRTModule:TensorRTRuntime"
             ):
-                if self.engine.has_implicit_batch_dimension:
-                    self.context.execute_async(
-                        batch_size, bindings, torch.cuda.current_stream().cuda_stream
-                    )
-                else:
-                    self.context.execute_async_v2(
-                        bindings, torch.cuda.current_stream().cuda_stream
-                    )
+                self.context.execute_async_v2(
+                    bindings, torch.cuda.current_stream().cuda_stream
+                )
 
             if len(outputs) == 1:
                 return outputs[0]

tests/py/dynamo/runtime/test_python_runtime.py

@@ -0,0 +1,81 @@
+import torch
+import torch_tensorrt
+from torch.testing._internal.common_utils import TestCase, run_tests
+
+
+class TestLowRankInputs(TestCase):
+    def test_0D_input(self):
+        class Tensor0DInput(torch.nn.Module):
+            def forward(self, x):
+                return x * 7
+
+        inputs = [
+            torch.tensor(
+                3,
+            )
+            .cuda()
+            .int(),
+        ]
+
+        fx_graph = torch.fx.symbolic_trace(Tensor0DInput())
+
+        # Validate that the results between Torch and Torch-TRT are similar
+        optimized_model = torch_tensorrt.compile(
+            fx_graph,
+            "torch_compile",
+            inputs,
+            min_block_size=1,
+            pass_through_build_failures=True,
+            use_python_runtime=True,
+        )
+        optimized_model_results = optimized_model(*inputs).detach().cpu()
+        torch_model_results = fx_graph(*inputs).detach().cpu()
+
+        max_diff = float(
+            torch.max(torch.abs(optimized_model_results - torch_model_results))
+        )
+        self.assertAlmostEqual(
+            max_diff,
+            0,
+            msg=f"0D-Tensor TRT outputs don't match with the original model.",
+        )
+        torch._dynamo.reset()
+
+    def test_1D_input(self):
+        class Tensor1DInput(torch.nn.Module):
+            def forward(self, x, y):
+                return (x + 7.1) / (y * 2.1)
+
+        inputs = [torch.rand((3, 1)).cuda(), torch.rand((3, 1)).cuda()]
+
+        fx_graph = torch.fx.symbolic_trace(Tensor1DInput())
+
+        # Validate that the results between Torch and Torch-TRT are similar
+        optimized_model = torch_tensorrt.compile(
+            fx_graph,
+            "torch_compile",
+            inputs,
+            min_block_size=1,
+            pass_through_build_failures=True,
+            use_python_runtime=True,
+        )
+        optimized_model_results = optimized_model(*inputs).detach().cpu()
+        torch_model_results = fx_graph(*inputs).detach().cpu()
+
+        max_diff = float(
+            torch.max(torch.abs(optimized_model_results - torch_model_results))
+        )
+        self.assertAlmostEqual(
+            max_diff,
+            0,
+            msg=f"1D-Tensor TRT outputs don't match with the original model.",
+        )
+
+        # Validate that the runtime moves cpu inputs to cuda
+        optimized_model(torch.rand((3, 1)), torch.rand((3, 1)))
+
+        torch._dynamo.reset()
+
+
+if __name__ == "__main__":
+    run_tests()