Error while running MNIST pre-built finn example

[nalnatsheh]

Hello everyone,

I am trying to run the pre-built finn example, but I am facing this error. Where to find this bitfile ?

[mmrahorovic]

Hi @n17a10d19e97r , thanks for your question!

From the latest FINN-examples release, we currently don't provide pre-built bitfiles for the Pynq-Z2 board. If you want to generate a bitfile for the TFC networks for the Pynq-Z2 board, you would have to do that manually. You can follow the steps here. More specifically, you would only need to edit the following line to zynq_platforms = ["Pynq-Z2"] and alveo_platforms=[] and then run the bnn-pynq/build.py build flow.
Alternatively, you could use a Pynq-Z1 board for which we have included pre-built bitfiles.

Hope this helps you further. Please let me know if you have any further questions!

[JJJWG]

Hi @mmrahorovic ,

I followed your guide to build the bitfiles manually for the Pynq-Z2 board as you described. After successfully building the bitfile for the Cybersecurity-FINN-Example I got the following output with the directories as shown in the Image:
~/finn-examples/build/cybersecurity-mlp/output_unsw_nb15-mlp-w2a2_Pynq-Z2:

How do I proceed from here and what should I do with the re-build bitfiles or the other output-files so that I can use the provided jupyter notebooks from FINN-Examples on my board? I am asking because I still have the same error as shown above, even tough I re-build the bitfile.

[mmrahorovic]

Hi @JJJWG,

Good question! The generated output files/folders are looking good from what I can tell.
In order to test the generated bitfile with one of the notebooks, you will have to make some changes to the notebook. At the moment, there is not a 'clean way' of doing this, so I will highlight the steps:

• In each notebook, a FINNExampleOverlay is created, which inherits the Pynq Overlay class and adds a few FINN-specific methods. When you create the FINNExampleOverlay, among others, a bitfile is downloaded on the FPGA. This is happening usually in the second code-block of each notebook. For the cybersecurity-MLP example: accel = models.mlp_w2a2_unsw_nb15().
• Looking at that function, you will need to change the filename variable to point to your newly generated bitfile. To do so, copy over the bitfile folder that was generated in the build process to your target board. Then, replace accel = models.mlp_w2a2_unsw_nb15() to the following:

from finn_examples.models import resolve_target_platform, get_driver_mode
from finn_examples.driver import FINNExampleOverlay
from qonnx.core.datatype import DataType

unsw_nb15_mlp_io_shape_dict = {
    "idt" : [DataType['BIPOLAR']],
    "odt" : [DataType['BIPOLAR']],
    "ishape_normal" : [(1, 600)],
    "oshape_normal" : [(1, 1)],
    "ishape_folded" : [(1, 15, 40)],
    "oshape_folded" : [(1, 1, 1)],
    "ishape_packed" : [(1, 15, 5)],
    "oshape_packed" : [(1, 1, 1)],
    "input_dma_name" : ['idma0'],
    "output_dma_name" : ['odma0'],
    "number_of_external_weights": 0,
    "num_inputs" : 1,
    "num_outputs" : 1,
}

def mlp_w2a2_unsw_nb15(target_platform=None):
    target_platform = resolve_target_platform(target_platform)
    driver_mode = get_driver_mode()
    filename = "/home/xilinx/bitfile/finn-accel.bit" # Path to the finn-accel.{bit,hwh} -- replace this with the location you copy the bitfile directory to
    fclk_mhz = 100.0
    return FINNExampleOverlay(
        filename,
        driver_mode,
        unsw_nb15_mlp_io_shape_dict,
        fclk_mhz=fclk_mhz
    )

accel = mlp_w2a2_unsw_nb15()

• You should now be good to go to run the rest of the notebook cells.

I have not tested this on my end, so please let me know if you run into any other problems!