Idan Ben Ami

Hi @YoshikiKato0220 , Looking at your log output, it appears that the error originates from 'torch.fx', which is used by MCT for converting a PyTorch model into graph. Usually it...

Hi @YoshikiKato0220, Thanks for your input. Please refer to this log output: File "/workspace/mmdetection/mmdet/models/detectors/base.py", line 91, in forward if mode == 'loss': This is a common problem "torch.fx" encounters when...

Hi @YoshikiKato0220 , Sorry for the delay. I see that this is also a "torch.fx' error from your log message, but I can't understand exactly what is the problem in...

Hi @ambitious-octopus, once I'm able to reproduce the issue in #1186, I'll start to debug this issue.

Hi @ambitious-octopus , We have found the root cause for this error. We noticed that your model performs operations on constants, such as ”to” and “mul” operations, which cause failures...

While avoiding operators like "to" seems to be correct for this model, we still need to address how to manage such issues. During torch FX, node names should be checked...

Hi @CYL0089 , Thank you for your feedback. Your suggestion is a valid solution to avoid this error. Another solution was suggested here: #1204 Thanks Idan

Our current recommendation is to avoid using the to() operator within the model's forward method before applying MCT. One problematic scenario occurs when to() is applied to constant values defined...