fp16的相关问题？如何设置fp16前向推理？

dlunion / tensorRTIntegrate

TensorRT ONNX Plugin、Inference、Compile

463 stars 114 forks source link

Closed BaofengZan closed 4 years ago

BaofengZan commented 4 years ago

https://github.com/dlunion/tensorRTIntegrate/blob/7543614a4ff5f6e022526a4e137d47f4e63f5f96/src/infer/trt_infer.cpp#L427-L429 这里再预处理图片时，强制规定type()是dtFloat，那如何进行半精度推理呢？
是在生成trtmodle时来指定fp16么？