Zhide-Zhou
Zhide-Zhou
> Hi, the redundant tag problem had been fixed. Please download the newest version. We will upload the evaluation code soon and inform you once done. Thank you very much...
> The fix has been merged into the main branch. Thanks! I have confirmed that this bug doesn't occur in the latest version of TVM.
> For relax.build() to work properly, it expects the model to be lowered to TensorIR before compilation. > > Before calling relax.build(), add this transformation: tvm_model = relax.transform.LowerToTensorIR()(tvm_model) > >...
> > > For relax.build() to work properly, it expects the model to be lowered to TensorIR before compilation. > > > Before calling relax.build(), add this transformation: tvm_model =...
> Well there are three options to resolve the issue Basically, the checker inside the ONNX is baked into each compiled wheel. > > I have tried first with the...
> Can you please share the onnx model? Sure, the onnx model is in the attachment 'testcase.zip'.
> Thanks for sharing the model. I checked the nodes and it seems like the first difference occurs after MaxPool. ORT thinks it should output [1,32,7,1] but in fact it...
> Can you attach the ONNX file? @LeoZDong Thank you for your reply! The ONNX file is in the following attachment. [testcase.zip](https://github.com/user-attachments/files/20622789/testcase.zip)
> Seems like the `ReduceMin` is getting a scalar instead of a 1D tensor. Can you check if setting `keepdims=1` for the `ReduceSumSquare` OP works? See [ONNX spec for ReduceSumSquare](https://github.com/onnx/onnx/blob/main/docs/Operators.md#attributes-86)...
> Could you try build from source with `--cmake_extra_defines onnxruntime_DEBUG_NODE_INPUTS_OUTPUTS=1`. Then you can set `ORT_DEBUG_NODE_IO_DUMP_INPUT_DATA=1` and `ORT_DEBUG_NODE_IO_DUMP_OUTPUT_DATA=1` before running test. It's able to see which node in CUDA provider generates...