Skip to content

Latest commit

 

History

History
95 lines (67 loc) · 4.64 KB

environment.md

File metadata and controls

95 lines (67 loc) · 4.64 KB
Raw

Offline Inference Environment Installation

This tutorial only covers the environment installation of MindOCR for offline inference on Atlas 300 series inference devices.

1. Version Matching Table

Please refer to the version matching table when setting up the inference environment. It is recommended to use MindSpore 2.2.14 for inference.

The version of Driver and Firmware is different for different chips. Please download the matched driver and firmware according to the CANN package version.

Now, we taking Atlas 300I Inference Card (Model: 3010) run on x86 CPU as an example, introduce version matching relationship. The following installation version is also introduced using this as an example.

MindSpore Driver Firmware CANN MindOCR
2.2.14 23.0.0 7.1.0.3.220 7.0.0.beta1 v0.4.0

Other MindSpore and Ascend software version matching please refer to MindSpore Install.

2. Ascend Environment Installation

There are two versions of the Ascend software package, the commercial edition and the community edition. The commercial edition is only for commercial customers and download is restricted; The community edition can be freely downloaded, and the following examples all use the community edition.

This example uses the Ascend package that comes with MindSpore 2.2.14, other MindSpore version please refer to Installing Ascend AI processor software package.

software version package name download
Driver 23.0.0 A300-3010-npu-driver_23.0.0_linux-x86_64.run link
Firmware 7.1.0.3.220 A300-3010-npu-firmware_7.1.0.3.220.run link
CANN nnae 7.0.0.beta1 Ascend-cann-nnae_7.0.0_linux-x86_64.run link
CANN kernels(Optional) 7.0.0.beta1 Ascend-cann-kernels-310p_7.0.0_linux.run link

Install

# Note: When installing a new machine, install the driver first and then the firmware.
# When the scenario of upgrade, install the firmware first and then the driver.
bash A300-3010-npu-driver_23.0.0_linux-x86_64.run --full
bash A300-3010-npu-firmware_7.1.0.3.220.run --full
bash Ascend-cann-nnae_7.0.0_linux-x86_64.run --install
bash Ascend-cann-kernels-310p_7.0.0_linux.run --install

pip uninstall te topi hccl -y
pip install sympy
pip install /usr/local/Ascend/nnae/latest/lib64/te-*-py3-none-any.whl
pip install /usr/local/Ascend/nnae/latest/lib64/hccl-*-py3-none-any.whl
reboot

Configure Environment Variables

source /usr/local/Ascend/nnae/set_env.sh

3. MindSpore Install

pip install mindspore==2.2.14

# Check version number, offline inference MindSpore only uses CPU
python -c "import mindspore;mindspore.set_context(device_target='CPU');mindspore.run_check()"

4. MindSpore Lite Install

software version package name download
Inference Toolkit 2.2.14 mindspore-lite-2.2.14-linux-{arch}.tar.gz link
Python Wheel 2.2.14 mindspore_lite-2.2.14-{python_version}-linux_{arch}.whl link

Unzip the inference toolkit and pay attention to setting environment variables:

tar -xvf mindspore-lite-2.2.14-linux-{arch}.tar.gz
cd mindspore-lite-2.2.14-linux-{arch}/
export LITE_HOME=${PWD}    # The actual path after extracting the tar package
export LD_LIBRARY_PATH=$LITE_HOME/runtime/lib:$LITE_HOME/runtime/third_party/dnnl:$LITE_HOME/tools/converter/lib:$LD_LIBRARY_PATH
export PATH=$LITE_HOME/tools/converter/converter:$LITE_HOME/tools/benchmark:$PATH

If interface with python, install the required whl package using pip.

pip install mindspore_lite-2.2.14-{python_version}-linux_{arch}.whl

5. MindOCR Install

git clone https://github.com/mindspore-lab/mindocr.git
cd mindocr
pip install -e .

Using '- e' to enter editable mode, help solve import issues.