Target platform 'DLXCKU5PE' is not supported for quantization.

Hi,

DLXCKU5PE is my self-generated deep learning bitstream with the datatype of int8. When I want to validate the quantized deep learning network on my FPGA platform.

An error occurred.

I wonder how to solve this problem. Or only offical evaluation boards support this function.

Answer 1

[Anjaneyulu Bairi]

Hi,

This error usually arises in Deep Learning HDL Toolbox when:

• The target FPGA platform you selected (DLXCKU5PE) is not officially supported by the quantization workflow in MATLAB/HDL Coder/Deep Learning HDL Toolbox.
• The platform is either custom or not included in the list of supported boards for quantized deployment.

Try to follow below steps which might helpful for you:

1. Check Supported Platforms

• Review the list of supported FPGA boards for quantized deep learning deployment in the https://in.mathworks.com/help/deep-learning-hdl/ug/supported-network-boards-and-tools.html
• If your board is not listed, quantized workflows (especially with int8) may not be possible out-of-the-box.

2. Custom Board Registration

• For custom boards, you may need to create a custom platform registration using the dlhdl.Target and dlhdl.Board classes, but quantization support may still be limited.

3. Try Float Deployment

• If quantized (int8) deployment is not supported, you may be able to deploy your network using single (floating point) precision instead.

I hope this helps!

[KH]

Thanks,

I am glad to receive your reply.

I have solved this problem by following the guide at

Quantize Network for FPGA Deployment - MATLAB & Simulink.

The Matlab can work with my platform.

However, the accuracy drops by roughly 4%. I’m currently looking into ways to recover this loss.