trt_profile_max_shapes not supported for ONNX-TRT backend

[ShuaiShao93]

Description
trt_profile_max_shapes is documented here, but it doesn't work

Triton Information
24.03

To Reproduce
Add this to onnx model config

optimization {
  execution_accelerators {
    gpu_execution_accelerator {
      name: "tensorrt"
      parameters {
        key: "max_workspace_size_bytes"
        value: "16000000000"
      }
      parameters {
        key: "precision_mode"
        value: "FP16"
      }
      parameters {
        key: "trt_engine_cache_enable"
        value: "1"
      }
      parameters {
        key: "trt_profile_max_shapes"
        value: "input_ids:8x256,attention_mask:8x256"
      }
      parameters {
        key: "trt_profile_min_shapes"
        value: "input_ids:1x256,attention_mask:1x256"
      }
      parameters {
        key: "trt_profile_opt_shapes"
        value: "input_ids:8x256,attention_mask:8x256"
      }
    }
  }
}

Start the triton server, it fails with Invalid argument: unknown parameter 'trt_profile_max_shapes' is provided for TensorRT Execution Accelerator;

Expected behavior
It should be supported as documentation

[rmccorm4]

Hi @ShuaiShao93,

I believe support for those parameters was just added recently here and should be included in the 24.04 release which was just released today. Please try it out and raise a new issue if that is not the case.