Ashwini Khade

From the error message it looks like it is unable to get the input "input_ids:0". Maybe some issue with input mapping not sure... needs investigation. How urgent is this?

@mayani-nv : This PR https://github.com/triton-inference-server/onnxruntime_backend/pull/42 to enable io binding should help with perf. Can you run your tests again once this is checked in? I have not done any perf...

onnxruntime is able to load this model and my test with random test data was also successful. This error is coming from ort backend in triton: https://github.com/triton-inference-server/onnxruntime_backend/blob/main/src/onnxruntime_utils.cc#L161 @deadeyegoodwin can triton...

Sequence in onnx is essentially a list. It can be a list of tensors, a list of list of tensors, list of map of tensors etc... In this case we...

Right now this op is not under consideration. As @linkerzhang suggested please send a PR for your proposal. @postrational and @gramalingam are OP SIG leads. Once you propose a PR...

In that case we can add a function op (which is composed of primitive ops) this way any runtime which has a specialized kernel can simply dispatch it and others...

> Thanks! If I understand it correctly, a function op will be a `FunctionProto` and the user can get the function body that contains the low-level ops. Right. The Op...

I will assign this to you so that we can track progress for this.

These 2 PRs clarify this behavior: https://github.com/onnx/onnx/pull/3532 https://github.com/onnx/onnx/pull/3575 Today we use this option " Define two versions of SoftmaxGrad, one targetting opset12 and one targetting opset13." Like we have discussed...

PR https://github.com/onnx/onnx/pull/3532 adds clarification for original questions.