Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

fix: bugs in TRT 10 upgrade #2832

Merged
merged 2 commits into from
May 18, 2024
Merged

fix: bugs in TRT 10 upgrade #2832

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
62673bd
fix bugs
zewenli98 May 14, 2024
1272940
change type
zewenli98 May 15, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 7 additions & 5 deletions py/torch_tensorrt/dynamo/conversion/_TRTInterpreter.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@
from typing import Any, Callable, Dict, List, NamedTuple, Optional, Sequence, Set

import numpy as np
import tensorrt as trt
import torch
import torch.fx
from torch.fx.node import _get_qualified_name
Expand All @@ -25,7 +26,6 @@
from torch_tensorrt.fx.observer import Observer
from torch_tensorrt.logging import TRT_LOGGER

import tensorrt as trt
from packaging import version

_LOGGER: logging.Logger = logging.getLogger(__name__)
Expand Down Expand Up @@ -313,8 +313,10 @@ def run(
)
timing_cache = self._create_timing_cache(builder_config, existing_cache)

engine = self.builder.build_serialized_network(self.ctx.net, builder_config)
assert engine
serialized_engine = self.builder.build_serialized_network(
self.ctx.net, builder_config
)
assert serialized_engine

serialized_cache = (
bytearray(timing_cache.serialize())
Expand All @@ -324,10 +326,10 @@ def run(
_LOGGER.info(
f"Build TRT engine elapsed time: {datetime.now() - build_engine_start_time}"
)
_LOGGER.info(f"TRT Engine uses: {engine.nbytes} bytes of Memory")
_LOGGER.info(f"TRT Engine uses: {serialized_engine.nbytes} bytes of Memory")

return TRTInterpreterResult(
engine, self._input_names, self._output_names, serialized_cache
serialized_engine, self._input_names, self._output_names, serialized_cache
)

def run_node(self, n: torch.fx.Node) -> torch.fx.Node:
Expand Down
8 changes: 4 additions & 4 deletions py/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@ class PythonTorchTensorRTModule(Module): # type: ignore[misc]

def __init__(
self,
engine: trt.ICudaEngine,
engine: trt.tensorrt.IHostMemory,
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Lets use the bytes interface (may be just changing the type annotation), but we use the same thing for the C++ runtime

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks. Currently, the engine actually saves serialized engine since TRT 10 changes API that uses builder.build_serialized_network. Do you recommend saving deserialized engines or serialized engines?

input_names: Optional[List[str]] = None,
output_names: Optional[List[str]] = None,
target_device: Device = Device._current_device(),
Expand Down Expand Up @@ -60,9 +60,9 @@ def _initialize(self) -> None:
self.engine = runtime.deserialize_cuda_engine(self.engine)
self.context = self.engine.create_execution_context()

assert (
self.engine.num_io_tensors // self.engine.num_optimization_profiles
) == (len(self.input_names) + len(self.output_names))
assert self.engine.num_io_tensors == (
len(self.input_names) + len(self.output_names)
)

self.input_dtypes = [
dtype._from(self.engine.get_tensor_dtype(input_name))
Expand Down
Loading