Incompatible with TensorRT 11

[Galasnow]

Katago v1.16.5 build failed with TensorRT 11.0.0 on Window Platform
According to the doc (https://docs.nvidia.com/deeplearning/tensorrt/latest/index.html), Some APIs used in "cpp/neuralnet/trtbackend.cpp" are deprecated:

"nvinfer1::ILayer",  "setPrecision"
"nvinfer1::IUnaryLayer", "setPrecision"
"nvinfer1::IScaleLayer",  "setPrecision"
"nvinfer1::IElementWiseLayer", "setPrecision"
"nvinfer1::IConvolutionLayer", "setPrecision"
"nvinfer1::IIdentityLayer", "setPrecision"
"nvinfer1::IActivationLayer", "setPrecision"
"nvinfer1::IConcatenationLayer", "setPrecision"
"nvinfer1::ITensor", "setType"
"nvinfer1::IBuilder", "platformHasFastFp16"
"nvinfer1::BuilderFlag", "kFP16"
"nvinfer1::BuilderFlag", "kPREFER_PRECISION_CONSTRAINTS"
"nvinfer1::NetworkDefinitionCreationFlag", "kEXPLICIT_BATCH"

[accrazed]

Can confirm it's also happening on Linux, re: pkg tensor 11.0.0.

Chewing through the repo and docs, pretty much all the precision / fp flag typing can be removed, in (mandatory) favor of their Model Optimizers' "AutoCast" converter.

This is also seemingly only available via cli or python api (O.O✿) . Seems like it'll be tedious adding a layer over the neuralnet/onnx* Builder to serialize via CLI/python

[GuangHui68]

官网提供的1.16.5，只匹配TRT 10.9吧？
你安装10.9就是了。并非最新的就是最合适的

[Galasnow]

官网提供的1.16.5，只匹配TRT 10.9吧？ 你安装10.9就是了。并非最新的就是最合适的

1. 迟早要更新到TRT 11.X的；
2. “只匹配TRT 10.9“不准确，v1.16.5在Cuda 13.3, TensorRT 10.16.1.11下编译通过，作者精力有限，只能测试若干版本。