Concrete-ML to tfhe-hpu-backend

Zama Concrete ML
1

Hi,
I am amazed at this framework, and working on my research using this tools.
I am recently trying to see if I can export the quantized trained mode done in concrete-ml in python, and run it using the tfhe-hpu-backend, so I can use the verilog simulation. But I cannot see the easy way to do it - seems like I need to take lots of detours.
Could you please let me know if there is any easier way of doing this? I saw the page about deployment, but it seems like it only stays at python and not reaching the hpu-backend.

Thank you, I would really appreciate any help.

2

Hi @Seoyoon_Jang,

There is no support of the HPU in Concrete ML as you might have guessed already.

I think the best approach for you is to re-implement the model in tfhe-rs.