Training a deep learning model for DPU compilation

TingShen_Kuo · February 24, 2021, 6:13am

Hi,I am now training a ssd mobile net v1 model on my host using pytorch.
I want to move it to ultra96 v2 board pynq using DPU, but i only see the tutorial using keras and tensorflow!

github.com

Xilinx/DPU-PYNQ/blob/master/host/train_mnist_model.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training a CNN for DPU compilation\n",
    "----"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Aim/s\n",
    "\n",
    "In this notebook we show how to train a simple Convolutional Neural Network (CNN)\n",
    "on the MNIST handwritten digit dataset for deployment on the DPU. We will cover:\n",
    "\n",
    "* Loading and pre-processing the imagenet dataset\n",

This file has been truncated. show original

I am wondering can I compile my model written by pytorch or there is some other tutorial for pytorch version?
Any advice would be appreciated!!
Thanks
Tim

rock · March 5, 2021, 4:59pm

Pytorch should be supported in DPU-PYNQ release v1.3.0. You can follow the steps on Vitis Ai to train your model and use DPU-PYNQ to deploy it.

Topic		Replies	Views
Can I import torch on PYNQ? Support	9	2135	July 24, 2020
DPU-PYNQ v1.3.0 release Announcements	2	1231	September 20, 2021
PYNQ + Vitis AI Libraries Support	2	366	August 2, 2021
General Question about training a CNN for DPU compilation Support	1	490	March 9, 2022
DPU-PYNQ now available for ZU+ and RFSoC devices Announcements	10	6578	May 21, 2022

Training a deep learning model for DPU compilation

Related topics