fix comments

Author: zewenli98

core/runtime/execute_engine.cpp

@@ -520,19 +520,20 @@ std::vector<at::Tensor> execute_engine(std::vector<at::Tensor> inputs, c10::intr
   if (compiled_engine->requires_output_allocator) { // engine requires OA
     if (cudagraphs_enabled) {
       TORCHTRT_THROW_ERROR(
-          "This module requires OutputAllocator which is not compatible with CUDA Graphs. Please disable CUDA Graphs.");
+          "The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs.");
     }
-    LOG_DEBUG("Using OutputAllocator in runtime.");
+    LOG_DEBUG("Using the dynamic allocator runtime mode.");
     return run_output_allocator();
   } else {
     if (compiled_engine->use_output_allocator_outputs) { // users call OA context manager
       if (cudagraphs_enabled) {
-        TORCHTRT_THROW_ERROR("Both CUDA Graphs and OutputAllocator are enabled. Please disable either one.");
+        TORCHTRT_THROW_ERROR(
+            "Both CUDA Graphs and dynamic output allocation are enabled, which are incompatible runtime modes. Please disable one of the two.");
       }
-      LOG_DEBUG("Using OutputAllocator in runtime.");
+      LOG_DEBUG("Using the dynamic allocator runtime mode.");
       return run_output_allocator();
     } else {
-      LOG_DEBUG("Using standard execution with cudagraphs=" << cudagraphs_enabled << ".");
+      LOG_DEBUG("Using the standard execution runtime mode with cudagraphs=" << cudagraphs_enabled << ".");
       return run_standard_execution();
     }
   }

core/runtime/runtime.h

@@ -16,7 +16,7 @@ namespace core {
 namespace runtime {
 
 using EngineID = int64_t;
-const std::string ABI_VERSION = "6";
+const std::string ABI_VERSION = "7";
 extern bool MULTI_DEVICE_SAFE_MODE;
 
 typedef enum {

py/torch_tensorrt/dynamo/runtime/_CudaGraphsTorchTensorRTModule.py

@@ -74,7 +74,7 @@ def __del__(self) -> None:
         if self.cudagraph:
             self.cudagraph.reset()
 
-    def set_output_allocator_outputs(self, enable: bool) -> None:
+    def set_use_output_allocator(self, enable: bool) -> None:
         self.use_output_allocator_outputs = enable
 
     def forward(self, *inputs: torch.Tensor) -> torch.Tensor | Tuple[torch.Tensor, ...]:

py/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.py

@@ -404,7 +404,7 @@ def create_output_tensors(self) -> List[torch.Tensor]:
     def set_pre_allocated_outputs(self, enable: bool) -> None:
         self.use_pre_allocated_outputs = enable
 
-    def set_output_allocator_outputs(self, enable: bool) -> None:
+    def set_use_output_allocator(self, enable: bool) -> None:
         self.use_output_allocator_outputs = enable
 
     def create_output_allocator(self) -> None:
@@ -683,21 +683,21 @@ def run_output_allocator() -> torch.Tensor | Tuple[torch.Tensor, ...]:
             if self.requires_output_allocator:  # engine requires OA
                 if self.cudagraphs_enabled:
                     raise RuntimeError(
-                        "This module requires OutputAllocator which is not compatible with CUDA Graphs. Please disable CUDA Graphs."
+                        "The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs."
                     )
-                logger.debug("Using OutputAllocator in runtime.")
+                logger.debug("Using the dynamic allocator runtime mode.")
                 return run_output_allocator()
             else:
                 if self.use_output_allocator_outputs:  # users call OA context manager
                     if self.cudagraphs_enabled:
                         raise RuntimeError(
-                            "Both CUDA Graphs and OutputAllocator are enabled. Please disable either one."
+                            "Both CUDA Graphs and dynamic output allocation are enabled, which are incompatible runtime modes. Please disable one of the two."
                         )
-                    logger.debug("Using OutputAllocator in runtime.")
+                    logger.debug("Using the dynamic allocator runtime mode.")
                     return run_output_allocator()
                 else:
                     logger.debug(
-                        f"Using standard execution with cudagraphs={self.cudagraphs_enabled}."
+                        f"Using the standard execution runtime mode with cudagraphs={self.cudagraphs_enabled}."
                     )
                     return run_standard_execution()
 

py/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.py

@@ -293,7 +293,7 @@ def set_extra_state(self, state: SerializedTorchTensorRTModuleFmt) -> None:
     def set_pre_allocated_outputs(self, enable: bool) -> None:
         self.engine.use_pre_allocated_outputs = enable
 
-    def set_output_allocator_outputs(self, enable: bool) -> None:
+    def set_use_output_allocator(self, enable: bool) -> None:
         self.engine.use_output_allocator_outputs = enable
 
     def forward(self, *inputs: Any) -> torch.Tensor | Tuple[torch.Tensor, ...]:

py/torch_tensorrt/runtime/_cudagraphs.py

@@ -81,7 +81,7 @@ def __enter__(self) -> torch.nn.Module:
                 and module.requires_output_allocator
             ):
                 raise RuntimeError(
-                    "There are converters that require Output Allocator. Please disable CUDA Graphs."
+                    "The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs."
                 )
             if "_run_on_acc" in name:
                 num_trt_module += 1

py/torch_tensorrt/runtime/_output_allocator.py

@@ -33,7 +33,7 @@ def __init__(
 
     def set_output_allocator_output(self, enable: bool) -> None:
         for mod in self.rt_mods:
-            mod.set_output_allocator_outputs(enable)
+            mod.set_use_output_allocator(enable)
 
     def __enter__(self) -> "_OutputAllocatorContextManager":
         # Enable output_allocator for TRT submodules

tests/py/dynamo/runtime/test_output_allocator_py.py

@@ -126,7 +126,7 @@ def test_combination_of_cg_and_oa(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="Both CUDA Graphs and OutputAllocator are enabled. Please disable either one.",
+            match="Both CUDA Graphs and dynamic output allocation are enabled, which are incompatible runtime modes. Please disable one of the two.",
         ):
             with torch_tensorrt.runtime.enable_cudagraphs(
                 compiled_model
@@ -136,7 +136,7 @@ def test_combination_of_cg_and_oa(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="Both CUDA Graphs and OutputAllocator are enabled. Please disable either one.",
+            match="Both CUDA Graphs and dynamic output allocation are enabled, which are incompatible runtime modes. Please disable one of the two.",
         ):
             with torch_tensorrt.runtime.enable_output_allocator(compiled_model):
                 with torch_tensorrt.runtime.enable_cudagraphs(
@@ -165,7 +165,7 @@ def test_cudagraphs_and_output_allocator(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="There are converters that require Output Allocator. Please disable CUDA Graphs.",
+            match="The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs.",
         ):
             with torch_tensorrt.runtime.enable_cudagraphs(
                 compiled_model
@@ -232,7 +232,7 @@ def test_combination_of_cg_and_oa(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="There are converters that require Output Allocator. Please disable CUDA Graphs.",
+            match="The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs.",
         ):
             with torch_tensorrt.runtime.enable_cudagraphs(
                 compiled_model
@@ -242,7 +242,7 @@ def test_combination_of_cg_and_oa(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="There are converters that require Output Allocator. Please disable CUDA Graphs.",
+            match="The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs.",
         ):
             with torch_tensorrt.runtime.enable_output_allocator(compiled_model):
                 with torch_tensorrt.runtime.enable_cudagraphs(
@@ -275,7 +275,7 @@ def test_cudagraphs_and_output_allocator(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="There are converters that require Output Allocator. Please disable CUDA Graphs.",
+            match="The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs.",
         ):
             with torch_tensorrt.runtime.enable_cudagraphs(
                 compiled_model
@@ -342,7 +342,7 @@ def test_combination_of_cg_and_oa(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="There are converters that require Output Allocator. Please disable CUDA Graphs.",
+            match="The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs.",
         ):
             with torch_tensorrt.runtime.enable_cudagraphs(
                 compiled_model
@@ -352,7 +352,7 @@ def test_combination_of_cg_and_oa(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="There are converters that require Output Allocator. Please disable CUDA Graphs.",
+            match="The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs.",
         ):
             with torch_tensorrt.runtime.enable_output_allocator(compiled_model):
                 with torch_tensorrt.runtime.enable_cudagraphs(
@@ -382,7 +382,7 @@ def test_cudagraphs_and_output_allocator(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="There are converters that require Output Allocator. Please disable CUDA Graphs.",
+            match="The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs.",
         ):
             with torch_tensorrt.runtime.enable_cudagraphs(
                 compiled_model
@@ -451,7 +451,7 @@ def test_combination_of_cg_and_oa(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="There are converters that require Output Allocator. Please disable CUDA Graphs.",
+            match="The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs.",
         ):
             with torch_tensorrt.runtime.enable_cudagraphs(
                 compiled_model
@@ -461,7 +461,7 @@ def test_combination_of_cg_and_oa(self, _, use_python_runtime):
 
         with pytest.raises(
             RuntimeError,
-            match="There are converters that require Output Allocator. Please disable CUDA Graphs.",
+            match="The model contains submodules that require a dynamic output allocator at runtime, which is incompatible with CUDA Graphs. Please disable CUDA Graphs.",
         ):
             with torch_tensorrt.runtime.enable_output_allocator(compiled_model):
                 with torch_tensorrt.runtime.enable_cudagraphs(