Closed PigletOS closed 2 years ago
Hello @PigletOS,
Thank you for the report. We are investigating the issue.
-Taylor
HI PigletOS
we are still working on this. We will keep you posted with updates.
Hi @PigletOS , We have implemented a workaround for the Framework error on Reshape_8. However, the model has very low ratio of compute to weight size as well as internal state size therefore the performance on Inferentia will be limited.
Environment torch 1.9.1 torch-neuron 1.9.1.2.0.392.0
Problem I tried to use torch_neuron to trace the StyleGAN2 model. But it failed with the error message "Error handling operator: Reshape_8 ". I also tried to trace the specific layers and found the error is related to the "ModulatedConv2d" layer. When it traced the "ModulatedConv2d" layer, it went into pdb mode.
Python Code and Logs torch_neuron_test.zip
Logs when tracing the StyleGAN2 model
Logs when tracing the ModulatedConv2d layer