From ae24156506cdcc07be43f66b5034c4ee43f396f3 Mon Sep 17 00:00:00 2001
From: Yuanle Liu <yuanlehome@163.com>
Date: Tue, 23 May 2023 15:01:43 +0800
Subject: [PATCH] Fix inference fp16 io (#54042)

* fix trt inference fp16 io

* fix inference fp16 io
---
 paddle/fluid/inference/tensorrt/convert/op_converter.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/paddle/fluid/inference/tensorrt/convert/op_converter.h b/paddle/fluid/inference/tensorrt/convert/op_converter.h
index 25af70f9d9c..018e3d2bb77 100644
--- a/paddle/fluid/inference/tensorrt/convert/op_converter.h
+++ b/paddle/fluid/inference/tensorrt/convert/op_converter.h
@@ -359,7 +359,7 @@ class OpConverter {
           platform::errors::InvalidArgument(
               "The output tensor in TensorRT subgraph should be LoDTensor"));
       nvinfer1::DataType out_dtype = FluidDataType2TRT(var->GetDataType());
-      if (engine->WithFp16() && !engine->WithInt8() &&
+      if (engine->precision() == phi::DataType::FLOAT16 &&
           out_dtype == nvinfer1::DataType::kFLOAT &&
           engine->EnableLowPrecisionIO()) {
         out_dtype = nvinfer1::DataType::kHALF;
-- 
GitLab