[Bug] W16A16 quantization, qdq_error is empty, but W8A8 is normal

microsoft / onnxruntime

ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator

https://onnxruntime.ai

MIT License

14.86k stars 2.94k forks source link

[Bug] W16A16 quantization, qdq_error is empty, but W8A8 is normal #21089

Open zccyman opened 5 months ago

zccyman commented 5 months ago

Describe Question:

when I use W16A16 quantization in main branch, I cann't get qdq_error and xmodel_err, because I found that qdq_error and xmodel_err are all empty. I eventually found that model.graph.value_info is same as the input model after the function "load_model_with_shape_infer", so I don't know how to fix it. I debug my code with "mobilenetv2-7.onnx".

yufenglee commented 5 months ago

@chenfucn, could you help take a look?

github-actions[bot] commented 4 months ago

This issue has been automatically marked as stale due to inactivity and will be closed in 30 days if no further activity occurs. If further support is needed, please provide an update and/or more details.