Registry / ai-ml / torch-tensorrt

torch-tensorrt

library2.12.0pypypiunverified

Torch-TensorRT is a compiler for PyTorch/TorchScript models targeting NVIDIA TensorRT. Version 2.12.0 supports Python 3.10+. It automatically optimizes models for inference on NVIDIA GPUs. Release cadence roughly quarterly.

ai-ml

pip install torch-tensorrt

github ↗homepage ↗

INSTALL

IMPORT

SIG · TORCH-TENSORRT

torch-tensorrt

ai-mlpythonv2.12.0

harness data pending

Install & Compatibility

Where this runs

No compatibility data collected yet for this library.

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

torch_tensorrt

✓ import torch_tensorrt

✗ import torchtrt

Old alias used before 1.4.0; no longer works

torch.jit

✓ import torch

TorchScript usage does not require separate import

Compile a PyTorch model using Torch-TensorRT with default settings.

import torch import torch_tensorrt model = torch.nn.Linear(10, 5).eval().cuda() inputs = torch.randn(1, 10).cuda() traced_model = torch.jit.trace(model, inputs) optimized_model = torch_tensorrt.compile( traced_model, inputs=[inputs], enabled_precisions={torch.float32} ) print(optimized_model(inputs))

Debug

Known issues

breakingWith Torch-TensorRT >= 2.0, the `torch.jit.script` path is deprecated and may be removed. Use `torch.jit.trace` or `torch_tensorrt.compile()` directly with torch.nn.Module.

fix

Use `torch.jit.trace` or pass module directly to `torch_tensorrt.compile` with `ir="torch_script"` or `ir="dynamo"`.

affects: >=2.0.0

deprecatedThe import `torchtrt` (underscore-less) is deprecated since version 1.4.0 and will be removed in a future release.

fix

Replace `import torchtrt` with `import torch_tensorrt`.

affects: >=1.4.0

gotchaPrecision specification changed. `enabled_precisions` argument takes a `set` of `torch.dtype` (e.g., `{torch.float32}`), not strings. In older versions you used strings like `'fp32'`.

fix

Use `torch.dtype` values in the set: `{torch.float32, torch.float16}`.

affects: >=2.0.0

gotchaWhen using TensorRT 10.x, Torch-TensorRT may require a specific minor version. Mismatch can cause runtime errors like 'CUDNN_STATUS_NOT_SUPPORTED'.

fix

Check compatibility matrix in official docs. Install `tensorrt` via `nvidia-tensorrt` pip package or use Docker containers.

affects: >=2.10.0

Upgrade

Version history

2.12.0latest on PyPI · released May 20, 2026

Audit

Dependencies

torchrequiredCore dependency; must be compatible version

tensorrtoptionalTensorRT runtime libraries (often via nvidia-tensorrt pip package)

Agent activity

19 hits · last 30 days

node

OpenAI (training)

Resources

githubhttps://github.com/pytorch/tensorrt.git ↗

homepagehttps://github.com/pytorch/tensorrt.git ↗