TVM EP and TensorRT EP do not support dynamic inputs

[farzanehnakhaee70]

Is your feature request related to a problem? Please describe.
Currently TVM EP and also TensorRT EP raises error while loading models with dynamic inputs. Is there any plan to support the models with dynamic inputs?

System information

• ONNX Runtime version (you are using): 1.9.0

[hariharans29]

CC: @stevenlix for the TRT question

@jywu-msft - Do you have the contact for the TVM EP part of the question ?

[jywu-msft]

TensorRT EP does support dynamic inputs. You may have to run symbolic shape inference script on the model. see https://onnxruntime.ai/docs/execution-providers/TensorRT-ExecutionProvider.html#shape-inference-for-tensorrt-subgraphs

[jywu-msft]

+@manashgoswami for pointer to TVM EP maintainers.

[manashgoswami]

The TVM ep doesn't support dynamic shape as the execution map is statically configured for the model in the TVM Runtime.

[farzanehnakhaee70]

Thanks a lot @manashgoswami

[yfirecanfly]

The TVM ep doesn't support dynamic shape as the execution map is statically configured for the model in the TVM Runtime.

Does onnxruntime tvm support targets such as cuda？