diff --git a/deployment/tensorrt/README.md b/deployment/tensorrt/README.md
index 6e7cb07a..c1a30c84 100644
--- a/deployment/tensorrt/README.md
+++ b/deployment/tensorrt/README.md
@@ -71,4 +71,7 @@ Here we will mainly discuss how to use the C++ interface, we recommend that you
                 --class_names {path/to/your/class/names}
    ```
 
+   Add minimal version of TensorRT
+   float32 需要8.2.0+ float16需要8.2.4+
+
    The above `yolort_trt` will determine if it needs to build the serialized engine file from ONNX based on the file suffix, and only do serialization when the argument `--model_path` given are with `.onnx` suffixes, all other suffixes are treated as the TensorRT serialized engine.