February 2022 Patch

New Features/Highlights

• Fixed a compiler bug for pt_yolox_TT100K_640_640_73G_2.0 model

• Fixed a quantizer bug in QAT in tensorflow 1.15 models

• Fixed a bug which gives low accuracy for TVM compiled inception_v3 model with DPUCADF8H(dpuv3int8)

• Fixed a bug in AI optimizer to get the best accuracy in run_evolutionary_search function

New Packages

• unilog-2.0.1-hea4fdf2_32.tar.bz2

• target_factory-2.0.1-h680af44_32.tar.bz2

• xir-2.0.1-py36h893bffd_32.tar.bz2

• xir-2.0.1-py37h893bffd_32.tar.bz2

• dpuv3-pycompiler-2.0.1-py36h3f44cd5_8.tar.bz2

• dpuv3-pycompiler-2.0.1-py37h3f44cd5_8.tar.bz2

• vai_optimizer_pytorch_gpu-2.0.1-py36h164a702_28.tar.bz2

• vai_q_tensorflow-1.15-py36hc48d084_2.tar.bz2

• vai_q_tensorflow_gpu-1.15-py36h91ed69b_2.tar.bz2

• vaic-2.0.1-py36hd51751d_11.tar.bz2

• vaic-2.0.1-py37hd51751d_11.tar.bz2

• xcompiler-2.0.1-py36hea4fdf2_32.tar.bz2

• xcompiler-2.0.1-py37hea4fdf2_32.tar.bz2

• xnnc-2.0.1-py36hd51751d_27.tar.bz2

• xnnc-2.0.1-py37hd51751d_27.tar.bz2

Installation

Download the packages from the link above. Apply the conda patch to the conda environment (Machine Learning framework) that you wish to update in this format

sudo conda install -n <CONDA_ENVIRONMENT> <URL or PATH to conda package>

For example, to update the vitis-ai-pytorch conda environment:

sudo conda install -n vitis-ai-pytorch https://www.xilinx.com/bin/public/openDownload?filename=unilog-2.0.1-hea4fdf2_32.tar.bz2

Make sure you install them in the order of unilog, target_factory, xir, xcompiler, xnnc, and use the proper version according to the python version installed in your current conda env.