Cpu memory optimization (#3602)

Author: cehongwang

py/torch_tensorrt/dynamo/_compiler.py

@@ -693,6 +693,7 @@ def compile(
 
     # Move the weights in the state_dict to CPU
     if offload_module_to_cpu:
+        deallocate_module(gm, delete_module=False)
         deallocate_module(exported_program.module(), delete_module=False)
         logger.info(
             "The PyTorch model was moved to the CPU to allocate all GPU memory to TensorRT. To retain the model on the GPU, set offload_module_to_cpu=False"

py/torch_tensorrt/dynamo/conversion/_TRTInterpreter.py

@@ -65,7 +65,7 @@ class UnsupportedOperatorException(RuntimeError):
 
 
 class TRTInterpreterResult(NamedTuple):
-    serialized_engine: bytes
+    engine: trt.ICudaEngine
     input_names: Sequence[str]
     output_names: Sequence[str]
     weight_name_map: Optional[dict[Any, Any]]
@@ -512,8 +512,7 @@ def _save_weight_mapping(self) -> None:
         _LOGGER.info("Building weight name mapping...")
         # Stage 1: Name mapping
         torch_device = to_torch_device(self.compilation_settings.device)
-        self.module.to(torch_device)
-        sd = self.module.state_dict()
+        sd = {k: v.to(torch_device) for k, v in self.module.state_dict().items()}
         weight_name_map: dict[str, Any] = {}
         weight_refit_map = self.ctx.weight_refit_map
         constant_mapping = {k: v for k, v in weight_refit_map.items() if v.size == 1}
@@ -592,13 +591,11 @@ def _save_weight_mapping(self) -> None:
         torch.cuda.empty_cache()
 
     @needs_refit  # type: ignore[misc]
-    def _insert_engine_to_cache(self, hash_val: str, serialized_engine: bytes) -> None:
+    def _insert_engine_to_cache(self, hash_val: str, engine: bytes) -> None:
+        serialized_engine = engine.serialize()
         # TODO: @Evan is waiting for TRT's feature to cache the weight-stripped engine
         # if not self.compilation_settings.strip_engine_weights:
         #     # set EXCLUDE_WEIGHTS flag to strip weights
-        #     runtime = trt.Runtime(TRT_LOGGER)
-        #     engine = runtime.deserialize_cuda_engine(serialized_engine)
-
         #     serialization_config = engine.create_serialization_config()
         #     serialization_config.set_flag(trt.SerializationFlag.EXCLUDE_WEIGHTS)
         #     serialized_engine = engine.serialize_with_config(
@@ -750,16 +747,15 @@ def run(
         self._create_timing_cache(
             builder_config, self.compilation_settings.timing_cache_path
         )
-        serialized_engine = self.builder.build_serialized_network(
+
+        cuda_engine = self.builder.build_engine_with_config(
             self.ctx.net, builder_config
         )
-        assert serialized_engine
+        assert cuda_engine
 
         _LOGGER.info(
             f"Build TRT engine elapsed time: {datetime.now() - build_engine_start_time}"
         )
-        _LOGGER.info(f"TRT Engine uses: {serialized_engine.nbytes} bytes of Memory")
-
         self.ctx.clear_cpu_weights_reference_holder()
 
         self._save_timing_cache(
@@ -772,14 +768,10 @@ def run(
             and self.compilation_settings.cache_built_engines
             and self.engine_cache is not None
         ):
-            self._insert_engine_to_cache(hash_val, serialized_engine)
-
-        with io.BytesIO() as engine_bytes:
-            engine_bytes.write(serialized_engine)
-            engine_str = engine_bytes.getvalue()
+            self._insert_engine_to_cache(hash_val, cuda_engine)
 
         return TRTInterpreterResult(
-            engine_str,
+            cuda_engine,
             self._input_names,
             self._output_names,
             self.weight_name_map,

py/torch_tensorrt/dynamo/conversion/_conversion.py

@@ -104,7 +104,7 @@ def convert_module(
         )
 
     return rt_cls(
-        serialized_engine=interpreter_result.serialized_engine,
+        cuda_engine=interpreter_result.engine,
         input_binding_names=list(interpreter_result.input_names),
         output_binding_names=list(interpreter_result.output_names),
         name=name,

py/torch_tensorrt/dynamo/lowering/passes/constant_folding.py

@@ -37,7 +37,9 @@ def constant_fold(
     # For TRT INetwork construction the constants are moved to CPU in get_attr call.
     for node, constant in cf.node_replacements.items():
         replace_node_with_constant(
-            gm, node, torch.nn.Parameter(constant, requires_grad=False)
+            gm,
+            node,
+            torch.nn.Parameter(constant.cpu().contiguous(), requires_grad=False),
         )
 
     erased_params = []

py/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.py

@@ -123,6 +123,7 @@ class PythonTorchTensorRTModule(Module):  # type: ignore[misc]
 
     def __init__(
         self,
+        cuda_engine: trt.ICudaEngine = None,
         serialized_engine: Optional[bytes] = None,
         input_binding_names: Optional[List[str]] = None,
         output_binding_names: Optional[List[str]] = None,
@@ -182,7 +183,19 @@ def __init__(
         # Unused currently - to be used by Dynamic Shape support implementation
         self.memory_pool = None
 
-        self.serialized_engine = serialized_engine
+        if cuda_engine:
+            assert isinstance(
+                cuda_engine, trt.ICudaEngine
+            ), "Cuda engine must be a trt.ICudaEngine object"
+            self.engine = cuda_engine
+        elif serialized_engine:
+            assert isinstance(
+                serialized_engine, bytes
+            ), "Serialized engine must be a bytes object"
+            self.engine = serialized_engine
+        else:
+            raise ValueError("Serialized engine or cuda engine must be provided")
+
         self.input_names = (
             input_binding_names if input_binding_names is not None else []
         )
@@ -204,7 +217,6 @@ def __init__(
             else False
         )
         self.settings = settings
-        self.engine = None
         self.weight_name_map = weight_name_map
         self.target_platform = Platform.current_platform()
         self.runtime_states = TorchTRTRuntimeStates(
@@ -219,7 +231,7 @@ def __init__(
         self.output_allocator: Optional[DynamicOutputAllocator] = None
         self.use_output_allocator_outputs = False
 
-        if self.serialized_engine is not None and not self.settings.lazy_engine_init:
+        if self.engine and not self.settings.lazy_engine_init:
             self.setup_engine()
 
     def get_streamable_device_memory_budget(self) -> Any:
@@ -260,13 +272,22 @@ def set_default_device_memory_budget(self) -> int:
         return self._set_device_memory_budget(budget_bytes)
 
     def setup_engine(self) -> None:
+
+        if isinstance(self.engine, trt.ICudaEngine):
+            pass
+        elif isinstance(self.engine, bytes):
+            runtime = trt.Runtime(TRT_LOGGER)
+            self.engine = runtime.deserialize_cuda_engine(self.engine)
+        else:
+            raise ValueError(
+                "Expected engine as trt.ICudaEngine or serialized engine as bytes"
+            )
+
         assert (
             self.target_platform == Platform.current_platform()
         ), f"TensorRT engine was not built to target current platform (target: {self.target_platform}, current: {Platform.current_platform()})"
 
         self.initialized = True
-        runtime = trt.Runtime(TRT_LOGGER)
-        self.engine = runtime.deserialize_cuda_engine(self.serialized_engine)
         if self.settings.enable_weight_streaming:
             self.set_default_device_memory_budget()
         self.context = self.engine.create_execution_context()
@@ -302,7 +323,7 @@ def _check_initialized(self) -> None:
             raise RuntimeError("PythonTorchTensorRTModule is not initialized.")
 
     def _on_state_dict(self, state_dict: Dict[str, Any], prefix: str, _: Any) -> None:
-        state_dict[prefix + "engine"] = self.serialized_engine
+        state_dict[prefix + "engine"] = self.engine
         state_dict[prefix + "input_names"] = self.input_names
         state_dict[prefix + "output_names"] = self.output_names
         state_dict[prefix + "platform"] = self.target_platform
@@ -317,7 +338,7 @@ def _load_from_state_dict(
         unexpected_keys: Any,
         error_msgs: Any,
     ) -> None:
-        self.serialized_engine = state_dict[prefix + "engine"]
+        self.engine = state_dict[prefix + "engine"]
         self.input_names = state_dict[prefix + "input_names"]
         self.output_names = state_dict[prefix + "output_names"]
         self.target_platform = state_dict[prefix + "platform"]

py/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.py

@@ -2,10 +2,12 @@
 
 import base64
 import copy
+import io
 import logging
 import pickle
 from typing import Any, List, Optional, Tuple, Union
 
+import tensorrt as trt
 import torch
 from torch_tensorrt._Device import Device
 from torch_tensorrt._enums import Platform
@@ -76,6 +78,7 @@ class TorchTensorRTModule(torch.nn.Module):  # type: ignore[misc]
 
     def __init__(
         self,
+        cuda_engine: Optional[trt.ICudaEngine | bytes] = None,
         serialized_engine: Optional[bytes] = None,
         input_binding_names: Optional[List[str]] = None,
         output_binding_names: Optional[List[str]] = None,
@@ -123,8 +126,22 @@ def __init__(
         """
         super(TorchTensorRTModule, self).__init__()
 
-        if not isinstance(serialized_engine, bytearray):
-            ValueError("Expected serialized engine as bytearray")
+        if serialized_engine:
+            assert isinstance(
+                serialized_engine, bytes
+            ), "Serialized engine must be a bytes object"
+            self.serialized_engine = serialized_engine
+
+        elif cuda_engine:
+            assert isinstance(
+                cuda_engine, trt.ICudaEngine
+            ), "Cuda engine must be a trt.ICudaEngine object"
+            serialized_engine = cuda_engine.serialize()
+            with io.BytesIO() as engine_bytes:
+                engine_bytes.write(serialized_engine)  # type: ignore
+                self.serialized_engine = engine_bytes.getvalue()
+        else:
+            raise ValueError("Serialized engine or cuda engine must be provided")
 
         self.input_binding_names = (
             input_binding_names if input_binding_names is not None else []
@@ -136,12 +153,11 @@ def __init__(
         self.hardware_compatible = settings.hardware_compatible
         self.settings = copy.deepcopy(settings)
         self.weight_name_map = weight_name_map
-        self.serialized_engine = serialized_engine
         self.engine = None
         self.requires_output_allocator = requires_output_allocator
 
         if (
-            serialized_engine
+            self.serialized_engine
             and not self.settings.lazy_engine_init
             and not self.settings.enable_cross_compile_for_windows
         ):