Fix inference fp16 io (#54042)

* fix trt inference fp16 io * fix inference fp16 io

Fix inference fp16 io (#54042)
* fix trt inference fp16 io * fix inference fp16 io
ae241565 · Yuanle Liu · GitHub · d89e0367 · ae241565
隐藏空白更改
内联并排

Showing with 1 addition and 1 deletion

paddle/fluid/inference/tensorrt/convert/op_converter.h paddle/fluid/inference/tensorrt/convert/op_converter.h +1 -1

未找到文件。
--- 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;