✨[Feature] Pre-compiled C++ Binaries for Windows

[wangzhenlin123]

Hi Team, Thank you for your work on Torch-TensorRT. I would like to inquire if there are any plans to provide official support and pre-compiled C++ binaries (LibTorchTRT) for the Windows platform. We have a use case that requires deploying optimized models on native Windows, and building from source is currently very difficult. Official support would be incredibly helpful for the community and for enterprise adoption on Windows systems.Is Windows C++ support on the project's roadmap?

Thanks for any information you can provide!

[narendasan]

Right now we aren't planning on a .zip release for Windows, perhaps in 2.9 but in the meantime you should be able to use the wheel files as a stand in. The built torchtrt.dll / torchtrt.lib files are available there and just need to be linked to run already exported torch-trt programs.

Example: https://github.com/pytorch/TensorRT/tree/main/examples/torchtrt_runtime_example

Can you provide information on what formats you use (e.g. ExportedProgram, AOT-Inductor, TorchScript)?

[wangzhenlin123]

Right now we aren't planning on a .zip release for Windows, perhaps in 2.9 but in the meantime you should be able to use the wheel files as a stand in. The built torchtrt.dll / torchtrt.lib files are available there and just need to be linked to run already exported torch-trt programs.

Example: https://github.com/pytorch/TensorRT/tree/main/examples/torchtrt_runtime_example

Can you provide information on what formats you use (e.g. ExportedProgram, AOT-Inductor, TorchScript)?

Thank you very much for the insightful response and for providing a viable path forward with the wheel files. We truly appreciate it.

I wanted to emphasize our use case: our company develops industrial inspection software, and the vast majority of our clients and deployment environments operate on Windows. Our core training and inference applications are built with C++, so having official, stable support for LibTorchTRT on Windows is a critical business need for us. It would significantly simplify our development and deployment pipeline.

We are eagerly looking forward to seeing official support, perhaps in a future release like 2.9 as you mentioned.Thank you again for your dedication to this project.

[lanluo-nvidia]

It is addressed in PR: https://github.com/pytorch/TensorRT/pull/3800 The windows zip will be avaiable from 2.9 release

[lanluo-nvidia]

@wangzhenlin123 Here is the zip folder libtorchtrt.zip ├── torch_tensorrt/ │ ├── include.zip
│ ├── include_core.zip
│ ├── lib.zip
│ ├── LICENSE │ └── BUILD

Let me know IF your use case can not be satisfied by this so that I can modify accordingly?