1、I'm sure the NPU driver and environment configuration is well because the GitHub tutorial example runs well.
2、Quantization and deploy on CPU completely
I use the default quantize configuration for u-net model and deployed the u net quantized model on CPU completely, the result is correct.
3、Deploy on the DPU
3.1 The session configuration is as following, and it seems nothing wrong with it.
3.2 The U-net model I use has some operators that is not supported by NPU.
3.3 When I run the model on DPU, the operators that DPU doesn't support is assigned to CPU and the compiler warns that I should specify the shape and the data type of the output tensor of this operation by
And the DPU warns DPU timeout: (Exception type: run: command error), Timeout layer name: [subgraph_/down1//maxpool_conv.0/MaxPool_output_0_DequantizeLinear_Output_vaip_71], Command failed to complete successfully (ERT_CMD_STATE_ERROR).
I guess that the operators DPU doesn't support is not defined in XIR and I should manually specify the shape and the data type.
4 Problems
So, is the operator DPU doesn't support cause the timeout?
Where should I write the command set data type to indicate the compiler.
Thanks for your help.
