训推一体化的AI飞桨套件：paddlex初探，还是不太顺利

paddlex本来是为了简化ai模型使用应运而生的，但是很长一段时间，都感觉它反而更难用。这回静下心来尝试一下

手册看这里：PaddleX 文档

repo:

PaddlePaddle/PaddleX: All-in-One Development Tool based on PaddlePaddle

先在星河社区进行实践。

安装

pip install "paddlex[base]"

果然，上来就碰到了问题

我在星河社区的项目里（是通过AI助手进来的项目，不是以前自己创立的项目），直接安装了paddlex之后，执行报错：

/bin/bash: paddlex: command not found

执行

到系统里，发现paddlex的执行文件在这个目录：

/home/aistudio/external-libraries/bin ,到这个目录去执行：
./paddlex –pipeline OCR \\
–input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \\
–use_doc_orientation_classify False \\
–use_doc_unwarping False \\
–use_textline_orientation False \\
–save_path ./output \\
–device gpu:0

发现卡住了

再执行一次，报错：

Pipeline prediction failed Traceback (most recent call last):   File "/home/aistudio/external-libraries/lib/python3.10/site-packages/paddlex/paddlex_cli.py", line 637, in main     pipeline_predict(   File "/home/aistudio/external-libraries/lib/python3.10/site-packages/paddlex/paddlex_cli.py", line 485, in pipeline_predict     for res in result:   File "/home/aistudio/external-libraries/lib/python3.10/site-packages/paddlex/inference/pipelines/_parallel.py", line 139, in predict     yield from self._pipeline.predict(   File "/home/aistudio/external-libraries/lib/python3.10/site-packages/paddlex/inference/pipelines/ocr/pipeline.py", line 357, in predict     det_results = list(   File "/home/aistudio/external-libraries/lib/python3.10/site-packages/paddlex/inference/models/base/predictor/base_predictor.py", line 281, in __call__     yield from self.apply(input, **kwargs)   File "/home/aistudio/external-libraries/lib/python3.10/site-packages/paddlex/inference/models/base/predictor/base_predictor.py", line 338, in apply     prediction = self.process(batch_data, **kwargs)   File "/home/aistudio/external-libraries/lib/python3.10/site-packages/paddlex/inference/models/text_detection/predictor.py", line 133, in process     batch_preds = self.infer(x=x)   File "/home/aistudio/external-libraries/lib/python3.10/site-packages/paddlex/inference/models/common/static_infer.py", line 298, in __call__     pred = self.infer(x)   File "/home/aistudio/external-libraries/lib/python3.10/site-packages/paddlex/inference/models/common/static_infer.py", line 261, in __call__     self.predictor.run() NotImplementedError: (Unimplemented) ConvertPirAttribute2RuntimeAttribute not support [pir::ArrayAttribute<pir::DoubleAttribute>]  (at /paddle/paddle/fluid/framework/new_executor/instruction/onednn/onednn_instruction.cc:116)

据说是框架不兼容导致的，临时解决方案是设置FLAGS_use_onednn=0

可以这样设置

import os
os.environ['FLAGS_use_onednn'] = '0'

对我们这个，这样设置

FLAGS_use_onednn=0 ./paddlex –pipeline OCR \\
–input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \\
–use_doc_orientation_classify False \\
–use_doc_unwarping False \\
–use_textline_orientation False \\
–save_path ./output \\
–device cpu

还是同样的报错

飞桨版本是3.3

pip show paddlepaddle Name: paddlepaddle Version: 3.3.0

在windows下安装测试

python -m pip install paddlex
python -m pip install paddlepaddle==3.3.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/

然后执行

paddlex –pipeline OCR –input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png –use_doc_orientation_classify False –use_doc_unwarping False –use_textline_orientation False –save_path ./output –device cpu

报错，说没有paddle，得，python3.14下好像还装不了飞桨…

第二天可以执行paddlex了

paddlex
Checking connectivity to the model hosters, this may take a while. To bypass this check, set `PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK` to `True`.
E:\\Programs\\Python\\Python314\\Lib\\site-packages\\langchain_core\\_api\\deprecation.py:27: UserWarning: Core Pydantic V1 functionality isn't compatible with Python 3.14 or greater.
from pydantic.v1.fields import FieldInfo as FieldInfoV1
No arguments provided. Displaying help information:
usage: Command-line interface for PaddleX. Use the options below to install plugins, run pipeline predictions, or start the serving application.
[-h] [–install [PLUGIN …]] [–no_deps] [–platform {github.com,gitee.com}] [-y] [–use_local_repos]
[–deps_to_replace DEPS_TO_REPLACE [DEPS_TO_REPLACE …]] [–pipeline PIPELINE] [–input INPUT]
[–save_path SAVE_PATH] [–device DEVICE] [–use_hpip] [–hpi_config HPI_CONFIG]
[–get_pipeline_config GET_PIPELINE_CONFIG] [–serve] [–host HOST] [–port PORT] [–paddle2onnx]
[–paddle_model_dir PADDLE_MODEL_DIR] [–onnx_model_dir ONNX_MODEL_DIR] [–opset_version OPSET_VERSION]

options:
-h, –help show this help message and exit

Install PaddleX Options:
–install [PLUGIN …]
Install specified PaddleX plugins.
–no_deps Install custom development plugins without their dependencies.
–platform {github.com,gitee.com}
Platform to use for installation (default: github.com).
-y, –yes Automatically confirm prompts and update repositories.
–use_local_repos Use local repositories if they exist.
–deps_to_replace DEPS_TO_REPLACE [DEPS_TO_REPLACE …]
Replace dependency version when installing from repositories.

Pipeline Predict Options:
–pipeline PIPELINE Name of the pipeline to execute for prediction.
–input INPUT Input data or path for the pipeline, supports specific file and directory.
–save_path SAVE_PATH
Path to save the prediction results.
–device DEVICE Device to run the pipeline on (e.g., 'cpu', 'gpu:0').
–use_hpip Use high-performance inference plugin.
–hpi_config HPI_CONFIG
High-performance inference configuration.
–get_pipeline_config GET_PIPELINE_CONFIG
Retrieve the configuration for a specified pipeline.

Serving Options:
–serve Start the serving application to handle requests.
–host HOST Host address to serve on (default: 0.0.0.0).
–port PORT Port number to serve on (default: 8080).

Paddle2ONNX Options:
–paddle2onnx Convert PaddlePaddle model to ONNX format.
–paddle_model_dir PADDLE_MODEL_DIR
Directory containing the PaddlePaddle model.
–onnx_model_dir ONNX_MODEL_DIR
Output directory for the ONNX model.
–opset_version OPSET_VERSION
Version of the ONNX opset to use.

调试

paddlex说没有paddle

Pipeline prediction failed Traceback (most recent call last):   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\paddlex_cli.py", line 637, in main     pipeline_predict(     ~~~~~~~~~~~~~~~~^         args.pipeline,         ^^^^^^^^^^^^^^     …<5 lines>…         **pipeline_args_dict,         ^^^^^^^^^^^^^^^^^^^^^     )     ^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\paddlex_cli.py", line 481, in pipeline_predict     pipeline = create_pipeline(         pipeline, device=device, use_hpip=use_hpip, hpi_config=hpi_config     )   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\pipelines\\__init__.py", line 168, in create_pipeline     pipeline = BasePipeline.get(pipeline_name)(         config=config,     …<5 lines>…         **kwargs,     )   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\utils\\deps.py", line 208, in _wrapper     return old_init_func(self, *args, **kwargs)   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\pipelines\\_parallel.py", line 113, in __init__     self._pipeline = self._create_internal_pipeline(config, self.device)                      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\pipelines\\_parallel.py", line 168, in _create_internal_pipeline     return self._pipeline_cls(            ~~~~~~~~~~~~~~~~~~^         config,         ^^^^^^^     …<5 lines>…         **self._init_kwargs,         ^^^^^^^^^^^^^^^^^^^^     )     ^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\pipelines\\ocr\\pipeline.py", line 76, in __init__     self.doc_preprocessor_pipeline = self.create_pipeline(                                      ~~~~~~~~~~~~~~~~~~~~^         doc_preprocessor_config         ^^^^^^^^^^^^^^^^^^^^^^^     )     ^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\pipelines\\base.py", line 140, in create_pipeline     pipeline = create_pipeline(         config=config,     …<5 lines>…         hpi_config=hpi_config,     )   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\pipelines\\__init__.py", line 168, in create_pipeline     pipeline = BasePipeline.get(pipeline_name)(         config=config,     …<5 lines>…         **kwargs,     )   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\utils\\deps.py", line 208, in _wrapper     return old_init_func(self, *args, **kwargs)   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\pipelines\\_parallel.py", line 113, in __init__     self._pipeline = self._create_internal_pipeline(config, self.device)                      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\pipelines\\_parallel.py", line 168, in _create_internal_pipeline     return self._pipeline_cls(            ~~~~~~~~~~~~~~~~~~^         config,         ^^^^^^^     …<5 lines>…         **self._init_kwargs,         ^^^^^^^^^^^^^^^^^^^^     )     ^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\pipelines\\doc_preprocessor\\pipeline.py", line 69, in __init__     self.doc_ori_classify_model = self.create_model(doc_ori_classify_config)                                   ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\pipelines\\base.py", line 106, in create_model     model = create_predictor(         model_name=config["model_name"],     …<7 lines>…         **kwargs,     )   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\models\\__init__.py", line 83, in create_predictor     return BasePredictor.get(model_name)(            ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^         model_dir=model_dir,         ^^^^^^^^^^^^^^^^^^^^     …<8 lines>…         **kwargs,         ^^^^^^^^^     )     ^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\models\\image_classification\\predictor.py", line 48, in __init__     super().__init__(*args, **kwargs)     ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\models\\base\\predictor\\base_predictor.py", line 169, in __init__     self._pp_option = self._prepare_pp_option(pp_option, device)                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\models\\base\\predictor\\base_predictor.py", line 415, in _prepare_pp_option     pp_option.device_type = device_info[0]     ^^^^^^^^^^^^^^^^^^^^^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\inference\\utils\\pp_option.py", line 186, in device_type     set_env_for_device_type(device_type)     ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^   File "E:\\Programs\\Python\\Python314\\Lib\\site-packages\\paddlex\\utils\\device.py", line 101, in set_env_for_device_type     import paddle ModuleNotFoundError: No module named 'paddle'

可能是飞桨不支持python3.14吧。换到python3.12环境下安装只用paddlex

提示WARNING: The script openai.exe is installed in 'E:\\py312\\Scripts' which is not on PATH.

11/15 [openai]  WARNING: The script openai.exe is installed in 'E:\\py312\\Scripts' which is not on PATH.   Consider adding this directory to PATH or, if you prefer to suppress this warning, use –no-warn-script-location.

两个python环境有点串啊

用这个方法

python -m paddlex –pipeline OCR –input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png –use_doc_orientation_classify False –use_doc_unwarping False –use_textline_orientation False –save_path ./output –device cpu

有新的报错：

报错OSError: cannot load library 'libsndfile.dll': error 0x7e

    from .readers import (   File "C:\\Users\\Admin\\AppData\\Roaming\\Python\\Python312\\site-packages\\paddlex\\inference\\utils\\io\\readers.py", line 33, in <module>     import soundfile   File "E:\\py312\\Lib\\site-packages\\soundfile.py", line 212, in <module>     _snd = _ffi.dlopen(_explicit_libname)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ OSError: cannot load library 'libsndfile.dll': error 0x7e

重新安装paddlepaddle==3.0.0

python -m pip install paddlepaddle==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/

有新的报错：

OSError: sndfile library not found using ctypes.util.find_library

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "<frozen runpy>", line 189, in _run_module_as_main
File "<frozen runpy>", line 148, in _get_module_details
File "<frozen runpy>", line 112, in _get_module_details
File "C:\\Users\\Admin\\AppData\\Roaming\\Python\\Python312\\site-packages\\paddlex\\__init__.py", line 49, in <module>
from .inference import create_pipeline, create_predictor
File "C:\\Users\\Admin\\AppData\\Roaming\\Python\\Python312\\site-packages\\paddlex\\inference\\__init__.py", line 16, in <module>
from .models import create_predictor
File "C:\\Users\\Admin\\AppData\\Roaming\\Python\\Python312\\site-packages\\paddlex\\inference\\models\\__init__.py", line 23, in <module>
from .anomaly_detection import UadPredictor
File "C:\\Users\\Admin\\AppData\\Roaming\\Python\\Python312\\site-packages\\paddlex\\inference\\models\\anomaly_detection\\__init__.py", line 15, in <module>
from .predictor import UadPredictor
File "C:\\Users\\Admin\\AppData\\Roaming\\Python\\Python312\\site-packages\\paddlex\\inference\\models\\anomaly_detection\\predictor.py", line 21, in <module>
from …common.batch_sampler import ImageBatchSampler
File "C:\\Users\\Admin\\AppData\\Roaming\\Python\\Python312\\site-packages\\paddlex\\inference\\common\\batch_sampler\\__init__.py", line 19, in <module>
from .image_batch_sampler import ImageBatchSampler
File "C:\\Users\\Admin\\AppData\\Roaming\\Python\\Python312\\site-packages\\paddlex\\inference\\common\\batch_sampler\\image_batch_sampler.py", line 23, in <module>
from …utils.io import PDFReader
File "C:\\Users\\Admin\\AppData\\Roaming\\Python\\Python312\\site-packages\\paddlex\\inference\\utils\\io\\__init__.py", line 16, in <module>
from .readers import (
File "C:\\Users\\Admin\\AppData\\Roaming\\Python\\Python312\\site-packages\\paddlex\\inference\\utils\\io\\readers.py", line 33, in <module>
import soundfile
File "E:\\py312\\Lib\\site-packages\\soundfile.py", line 212, in <module>
_snd = _ffi.dlopen(_explicit_libname)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
OSError: cannot load library 'libsndfile.dll': error 0x7e

AI说

遇到的报错信息 OSError: sndfile library not found using ctypes.util.find_library 是由于系统缺少 ‌libsndfile‌ 动态链接库导致的。这个错误在使用 PaddleX 的某些功能（如音频处理、OCR 中涉及 PDF/音频读取）时会触发。

PaddleX 在加载某些模块（如 PDF 或音频读取）时依赖 soundfile 库，而 soundfile 又依赖系统中的 libsndfile.dll（Windows 下）或 libsndfile.so（Linux 下）。当前环境未找到该库，因此报错。

安装 libsndfile（推荐）‌

1. 下载 libsndfile

前往 libsndfile 官网 或 GitHub 发布页 下载适用于 Windows 的 DLL 文件（例如 libsndfile-1.0.31-64bit.zip）。

2. 解压并复制 DLL 文件

将下载的 libsndfile-1.0.31.dll 文件复制到以下位置之一：

• Python 安装目录下的 DLLs 文件夹（如 E:\\py312\\DLLs）
• 系统 PATH 环境变量中的某个目录（如 C:\\Windows\\System32）

到github下载：Releases · libsndfile/libsndfile

解压，把sndfile文件放到dlls目录，不管用。我发现它需要的文件是libsndfile.dll,而解开之后的文件是sndfile.dll

使用pip安装

python -m pip install libsndfile

还是不行。