How to quantize specific layers at OpenVINO Workbench?

[anvarganiev07]

I have a trained onnx model that needs to be quantized to INT8. But I want my last fully connected layers are still in FP32 or FP16. So how can I choose specific layers to quantize (or not to quantize)?

PS when I was working with NNCF, I just use parameter ignored_scopes. Maybe is there something similar here at Workbench?

[Wan_Intel]

Hi Anvarganiev07,

Thank you for reaching out to us.

Referring to Optional parameters in DefaultQuantization Parameter, you may use "ignored" parameter to exclude nodes or operation types from optimization. For example, you may refer to the following links:

• Post-Training Quantization Best Practices
• Full specification for AccuracyAwareQuantization algorithm
• Full specification for DefaultQuantization algorithm

Hope it helps.

Regards,

Wan

[Wan_Intel]

Hi Anvarganiev07,

Thanks for your question.

This thread will no longer be monitored since we have provided a suggestion. 

If you need any additional information from Intel, please submit a new question.

Regards,

Wan