Installation

Stable versions of Torch-TensorRT are published on PyPI

pip install torch-tensorrt

Nightly versions of Torch-TensorRT are published on the PyTorch package index

pip install --pre torch-tensorrt --index-url https://download.pytorch.org/whl/nightly/cu128

Torch-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

Quickstart

Option 1: torch.compile

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!

Option 2: Export

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).

Step 1: Optimize + serialize

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)

Step 2: Deploy

Deployment in PyTorch:

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)

Deployment in C++:

#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});

Further resources

• 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

Note: Refer NVIDIA L4T PyTorch NGC container for PyTorch libraries on JetPack.

Dependencies

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 8.1.1
• Libtorch 2.8.0.dev (latest nightly)
• CUDA 12.8 (CUDA 12.6 on Jetson)
• TensorRT 10.11 (TensorRT 10.3 on Jetson)

Deprecation Policy

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.

Contributing

Take a look at the CONTRIBUTING.md

License

The Torch-TensorRT license can be found in the LICENSE file. It is licensed with a BSD Style licence