scragnog/HOT-Step-CPP-TensorRT

TensorRT runtime for HOT-Step CPP

Redistributable NVIDIA TensorRT 10.16.1.11 (CUDA 13) Windows runtime libraries, downloaded on demand by the HOT-Step CPP Model Manager so the app can build and run its native MM3 DiT engine. They are distributed as part of the HOT-Step CPP application under the NVIDIA TensorRT Software License Agreement (the runtime .dll files are the SLA's designated distributable portion). Copyright and version information inside the files is unchanged. Not a standalone SDK: use them only with HOT-Step CPP.

tensorrt-10.16.1.11-headers.zip holds the TensorRT C++ headers, which NVIDIA publishes under Apache-2.0 at https://github.com/NVIDIA/TensorRT; HOT-Step's CI compiles against it.

File	Purpose
nvinfer_10.dll	inference runtime (required)
nvonnxparser_10.dll	ONNX parser, used once to build the engine (required)
nvinfer_builder_resource_smNN_10.dll	builder resources; you need the one matching your GPU (sm120 = RTX 50, sm89 = RTX 40, sm86 = RTX 30, sm75 = RTX 20, sm80/90/100 = datacenter)
nvinfer_builder_resource_ptx_10.dll	fallback for GPUs without a matching resource file