❓ [Question] Using torch-trt to test bert's qat quantitative model

[lixiaolx]

❓ Question

When using torch-trt to test Bert's qat quantization ( https://zenodo.org/record/4792496#.YxGrdRNBy3J ) model, I encountered many FakeTensorQuantFunction nodes in the pass, and at the same time triggered many nodes that could not convert TRT, and split the graph into many subgraphs

question:

1. Can you tell me how to explain the nodes that appear in the pass, and how to explain the symbols (^) in front of these nodes?
2. How can these quantization nodes be converted into qat nodes corresponding to torch-trt（ https://github.com/pytorch/TensorRT/blob/master/core/conversion/converters/impl/quantization.cpp ）?

[ncomly-nvidia]

@peri044 What nodes are inserted by the PyT QAT Toolkit? Are they FakeTensorQuantFunction layers?

[github-actions[bot]]

This issue has not seen activity for 90 days, Remove stale label or comment or this will be closed in 10 days

[github-actions[bot]]

This issue has not seen activity for 90 days, Remove stale label or comment or this will be closed in 10 days