PaddleOCR/latest/version3.x/other_devices_support/paddlepaddle_install_NPU

[giscus[bot]]

PaddleOCR/latest/version3.x/other_devices_support/paddlepaddle_install_NPU

Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)

https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/other_devices_support/paddlepaddle_install_NPU.html

[ydcloud-dy]

[root@k8s-node03 paddle]# pip install paddlepaddle==3.0.0.dev20250527 -i https://www.paddlepaddle.org.cn/packages/nightly/cpu
ERROR: ld.so: object '/usr/lib/aarch64-linux-gnu/libgomp.so.1' from LD_PRELOAD cannot be preloaded (cannot open shared object file): ignored.

ERROR: ld.so: object '/usr/lib/aarch64-linux-gnu/libgomp.so.1' from LD_PRELOAD cannot be preloaded (cannot open shared object file): ignored.
Looking in indexes: https://www.paddlepaddle.org.cn/packages/nightly/cpu
ERROR: Could not find a version that satisfies the requirement paddlepaddle==3.0.0.dev20250527
ERROR: No matching distribution found for paddlepaddle==3.0.0.dev20250527
WARNING: You are using pip version 20.3.3; however, version 24.0 is available.
You should consider upgrading via the '/usr/bin/python3 -m pip install --upgrade pip' command.

下载安装 wheel 安装包的时候找不到包啊

[xgySTATISICT]

支持910A么？支持的话，panddlepaddle和paddleocr分别要使用什么版本？

[Omega-wu]

在安装完wheel之后出现了找不到npu的错误，但是npu-smi info可以看到显卡信息，环境变量也设置了。错误信息是这样的
server:/work# python -
ot@ubuntu-s
port paddle; paddle.utils.run
che
local/lib/python3.10/dist-packages/paddle_custom_device
2 11:58:10.204818
init.cc:238]
[CUSTOM_DEVICE_ROOT]=/usr
121:
I1022
121 init.cc:146] rry 1oading custom device 1ibs from: [/usr/local/lib/python3.10/dist-packages/paddle_custom devicel
ENV
121 custom_device_load.cc:51]
11022 11:58:10.204907
2 11:58:11.517750
Succeed in loading cu
runtime in lib: /usr/local/lib/python3.10/dist-packages/paddle_custom_device/libpaddle-custom-npu.s0
121 custom_device_1oad.cc:58] Skipped lib [/usr/local/lib/python3.10/dist-packages/paddle_custom_device/libpaddle-custom-npu.so]: no custom engine Plugin symbol in
I1022
11022 11:58:11.517858
this lib.
11022 11:58:11.523918
121 custom_kernel.cc:68] Succeed in loading 359 custom kernel(s) from loaded lib(s), will be used like native ones.
11022 11:58:11.524173
121 init.cc:158] Finished in LoadcustomDevice with libs_path: [/usr/local/lib/python3.10/dist-packages/paddle_custom_device]
11022 11:58:11.524269
121 init.cc:244] CustomDevice: mpu, visible devices count:
/usr/local/lib/python3.10/dist-packages/paddle/base/framework.py:854: UserWarning: You are using CUSTOM DEVICE version Paddle, but your custom device is not set properly. CPU device will
e used by default.
warnings.warn(
Running verify PaddlePaddle program
E1022 11:58:12.194439
1 place.cc:230] Cannot use npu because there is no npu detected on your machine.
Traceback (most recent call last):
121
File "", line 1, in
File "/usr/local/lib/python3.10/dist-packages/paddle/utils/install_check.py", line 273, in run_check
run static single (use_cuda, use xpu, use custom, custom device_name)
File "/usr/1ocal/lib/python3.10/dist-packages/paddle/utils/install_check.py", line 147, in run_static single
place = paddle.CustomPlace (custom_device name, 0)
ValueError:
(InvalidArg
Please
check. (at /paddle/paddle/fluid/pybind/place.cc:235)
rong place,

[sdfu9cjuhdf]

卸载当前版本

pip uninstall paddlepaddle paddle-custom-npu

安装最新的开发版本

python -m pip install paddlepaddle==3.0.0.dev20250610 -i https://www.paddlepaddle.org.cn/packages/nightly/cpu
python -m pip install paddle-custom-npu==3.0.0.dev20250610 -i https://www.paddlepaddle.org.cn/packages/nightly/npu

NPU要用这个版本，我是910B3.理论上910B和300的都可以。

[yhz0120]

1. 我用的300I Pro 推理卡
2. 相关版本使用如下
   paddle-custom-npu 0.0.0 # 基于PaddleCustomDevice 的develop分支自己编译安装的
   paddleocr 3.3.0
   paddlepaddle 3.2.0
   paddlex 3.3.6
3. 健康检查可以通过
4. 问题如下
   paddleocr ocr --device cpu 可以识别图片并输出文本内容
   paddleocr ocr --device npu 输出的文件内容为空，整个过程中可以观察到npu卡有内存波动

请帮忙分析下 # 2025.10.30