Dumping weights.

[Ritwika_C_Intel]

How can we dump the intermediate layer weights in the latest openvino?

How can we convert the weights to Int7? Where in the code can we make changes for the same?

[IntelSupport]

Hi Ritwika,

Thanks for reaching out.

Could you explain in detail what are you trying to accomplish? Are you trying to add a custom layer to a model? You could refer to this Custom Operations Guide to learn more regarding the operation needed. Else, if you trying to convert the weight model into Intermediate Representation(IR) of OpenVINO, refer to Converting a Model Using General Conversion Parameters documentation.

Regards,

Aznie

[Ritwika_C_Intel]

Hi Aznie,

Thank you for your reply.

I am not trying to add a custom layer. I am trying to achieve the following:

1. Write the weights for all the layers in a text file.

2. Convert the weights to INT 7.

In 2019 version of openvino, there was a file called cnn_network_int8_normalizer.cpp. By making some additions to the code there, the weights were converted to Int7. However that file do not exist in the latest openvino. I want to know a suitable place in the code where I can make modifications to convert it to Int7.

Thanks,

Regards,

Ritwika

[IntelSupport]

Hi Ritwika,

Thank you for your patient. OpenVINO only supports the lowest up to 1-byte integer format, which is I8 format or U8 format. Based on this documentation, you can only use mo.py to convert the model to IR format with the precision shown in the documents.

For the cnn_network_int8_normalizer.cpp, OpenVINO no longer supports cnn_network_int8_normalizer and we no longer publish this code anymore.

Regards,

Aznie

[IntelSupport]

Hi Ritwika,

This thread will no longer be monitored since we have provided a solution. If you need any additional information from Intel, please submit a new question.

Regards,

Aznie