Torch-TensorRT brings the power of TensorRT to PyTorch. Accelerate inference latency by up to 5x compared to eager execution in just one line of code.
Stable versions of Torch-TensorRT are published on PyPI
pip install torch-tensorrtNightly versions of Torch-TensorRT are published on the PyTorch package index
pip install --pre torch-tensorrt --index-url https://download.pytorch.org/whl/nightly/cu124Torch-TensorRT is also distributed in the ready-to-run NVIDIA NGC PyTorch Container which has all dependencies with the proper versions and example notebooks included.
For more advanced installation methods, please see here
You can use Torch-TensorRT anywhere you use torch.compile:
import torch
import torch_tensorrt
model = MyModel().eval().cuda() # define your model here
x = torch.randn((1, 3, 224, 224)).cuda() # define what the inputs to the model will look like
optimized_model = torch.compile(model, backend="tensorrt")
optimized_model(x) # compiled on first run
optimized_model(x) # this will be fast!If you want to optimize your model ahead-of-time and/or deploy in a C++ environment, Torch-TensorRT provides an export-style workflow that serializes an optimized module. This module can be deployed in PyTorch or with libtorch (i.e. without a Python dependency).
import torch
import torch_tensorrt
model = MyModel().eval().cuda() # define your model here
inputs = [torch.randn((1, 3, 224, 224)).cuda()] # define a list of representative inputs here
trt_gm = torch_tensorrt.compile(model, ir="dynamo", inputs=inputs)
torch_tensorrt.save(trt_gm, "trt.ep", inputs=inputs) # PyTorch only supports Python runtime for an ExportedProgram. For C++ deployment, use a TorchScript file
torch_tensorrt.save(trt_gm, "trt.ts", output_format="torchscript", inputs=inputs)import torch
import torch_tensorrt
inputs = [torch.randn((1, 3, 224, 224)).cuda()] # your inputs go here
# You can run this in a new python session!
model = torch.export.load("trt.ep").module()
# model = torch_tensorrt.load("trt.ep").module() # this also works
model(*inputs)#include "torch/script.h"
#include "torch_tensorrt/torch_tensorrt.h"
auto trt_mod = torch::jit::load("trt.ts");
auto input_tensor = [...]; // fill this with your inputs
auto results = trt_mod.forward({input_tensor});- Up to 50% faster Stable Diffusion inference with one line of code
- Optimize LLMs from Hugging Face with Torch-TensorRT [coming soon]
- Run your model in FP8 with Torch-TensorRT
- Tools to resolve graph breaks and boost performance [coming soon]
- Tech Talk (GTC '23)
- Documentation
| Platform | Support |
|---|---|
| Linux AMD64 / GPU | Supported |
| Windows / GPU | Supported (Dynamo only) |
| Linux aarch64 / GPU | Native Compilation Supported on JetPack-4.4+ (use v1.0.0 for the time being) |
| Linux aarch64 / DLA | Native Compilation Supported on JetPack-4.4+ (use v1.0.0 for the time being) |
| Linux ppc64le / GPU | Not supported |
Note: Refer NVIDIA L4T PyTorch NGC container for PyTorch libraries on JetPack.
These are the following dependencies used to verify the testcases. Torch-TensorRT can work with other versions, but the tests are not guaranteed to pass.
- Bazel 6.3.2
- Libtorch 2.5.0.dev (latest nightly) (built with CUDA 12.4)
- CUDA 12.4
- TensorRT 10.3.0.26
Deprecation is used to inform developers that some APIs and tools are no longer recommended for use. Beginning with version 2.3, Torch-TensorRT has the following deprecation policy:
Deprecation notices are communicated in the Release Notes. Deprecated API functions will have a statement in the source documenting when they were deprecated. Deprecated methods and classes will issue deprecation warnings at runtime, if they are used. Torch-TensorRT provides a 6-month migration period after the deprecation. APIs and tools continue to work during the migration period. After the migration period ends, APIs and tools are removed in a manner consistent with semantic versioning.
Take a look at the CONTRIBUTING.md
The Torch-TensorRT license can be found in the LICENSE file. It is licensed with a BSD Style licence
