Error in PYNQ-DPU loading "xmodel"

Support
1

Hi, I´m trying to upload the model compiled from train_mnist_classifier example. I am using PYNQ 2.6 , Vivado 2020.1 and PYNQ-DPU 1.2.
When I compile the model, I get a “.xmodel” file instead of “.elf”. Then, I uploaded the model and the “dpu.bit” files to the board, but when I try to run this code in Jupyter Notebook:

from pynq_dpu import DpuOverlay
overlay = DpuOverlay(“…/mymodel/dpu.bit”)
overlay.load_model(“…/mymodel/mnist_classifier.xmodel”)

appears this error:

RuntimeError Traceback (most recent call last)
in ()
1 from pynq_dpu import DpuOverlay
2 overlay = DpuOverlay(“…/mymodel/dpu.bit”)
----> 3 overlay.load_model(“…/mymodel/mnist_classifier_org.xmodel”)

/usr/local/lib/python3.6/dist-packages/pynq_dpu/dpu.py in load_model(self, model)
211
212 if not model.endswith(“.elf”):
→ 213 raise RuntimeError(“Currently only elf files can be loaded.”)
214 else:
215 if self.runtime == ‘dnndk’:

RuntimeError: Currently only elf files can be loaded.

Any suggestions?
Regards

1 Like
2

xmodel is for Vitis AI 1.3. Check the README here:
"
For the GPU accelerated docker image:

./docker_run.sh xilinx/vitis-ai:1.2.82
From version 1.3 .xmodel files are generated instead of .elf
"

1 Like
3

Is there any plan to support this soon? If not, is there a possible way of creating the .elf file based on vitis ai 1.3?

4

It is already supported in DPU-PYNQ v1.3.0

5

So s you mean I can natively generate the .elf file even for vitis ai 1.3 while using DPU PYNQ repo? what about the .xmodel files, are they supported by PYQN ?
As far as I know PYNQ 2.6 doesn’t support .xmodel file (I just update the dpu-pynq version of my ultra96v2) and I’m still not able to load them, because is not an elf file.

6

Sorry for the confusion, just for documenting, after upgrading my pynq version I had to update pynq-dpu to its latest version, after that I was able to load .xmodel files.

1 Like