Skip to content

Infer Brevitas CNV model on Non-PYNQ board #76

@abedbaltaji

Description

@abedbaltaji

Hello,

I have generated a stitched IP using FINN builder of the Brevitas cnv-w2a2 pre-trained model using the notebook under "finn/notebookes/advanced/4_advanced_builder_settings.ipynb", then implemented it on a FPGA target board (NOT PYNQ) which has been connected to communication blocks to read/write to the finn design.

My question is that how should I prepare/feed the input data concerning its folding shape and packaging its inner most dimension in order to obtain a ".dat" file? I got inspired from the data preparation function "custom_step_gen_tb_and_io", that I found in the "finn/tutorials/fpga_flow/build.py" script, but it is for the MNIST classification (tfc-w1a1), but I did not succeed to adapt it for this model (CNV).

Your help would be appreciated!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions