nuitka打包环境下无法将动态图训练模型转换为静态图推理模型

[nihaosha]

🔎 Search before asking

• I have searched the PaddleOCR Docs and found no similar bug report.
• I have searched the PaddleOCR Issues and found no similar bug report.
• I have searched the PaddleOCR Discussions and found no similar bug report.

🐛 Bug (问题描述)

采用nuitka打包paddleocr后，进行推理任务执行，报错如下：OSError: could not get source code

utils\analyze\ocr.py", line 1499, in _regenerate_recognittion_inference_model paddleocr\ppocr\utils\export_model.py", line 514, in export paddleocr/ppocr\utils\export_model.py", line 388, in export_single_model ‚decorator.py", line 235, in fun „paddle\fluid\wrapped_decorator.py", line 26, in __impl__ paddle\fluid\dygraph\jit.py", line 649, in wrapper ‚decorator.py", line 235, in fun paddle\fluid\wrapped_decorator.py", line 26, in __impl__ paddle\fluid\dygraph\base.py", line 67, in __impl__ „paddle\fluid\dygraph\jit.py", line 927, in save paddle\fluid\'dygraph\dygraph_to_static\program_translator.py", line 577, in concrete_program_specify_input_spec paddle\fluid\dygraph\dygraph_to_static\program_transslator.py", line 485, in get_concrete_program paddle\fluid\dygraph\dygraph_to_static\program_transslator.py", line 955, in _getitem_ paddle\fluid\dygraph\dygraph_to_static\program_translator.py", line 939, in _build_once ‚decorator.py", line 235, in fun paddle\fluid\wrapped_decorator.py", line 26, in __impl__ .paddle\fluid\dygraph\base.py", line 67, in __impl__ paddle\fluid\dygraph\dygraph_to_static\program_transslator.py", line 848, in from_func_spec paddle\fluid\dygraph\dygraph_to_static\program_translatorr.py", line 142, in convert_to_static paddle\fluid\dygraph\dygraph_to_static\program_transslator.py", line 77, in convert_with_cache paddle\fluid\dygraph\dygraph_to_static\program_translatoxr.py", line 104, in _convert paddle\fluid\dygraph\dygraph_to_static\utils.py", line 703, in func_to_source_code inspect.py", line 1121, in getsourcelines inspect.py", line 958, in findsource

转换原因分析：
直接原因：inspect.getsourcelines/inspect.findsource 这类 Python 标准库的源码检查工具，只能获取有源码文件的函数/类的源代码。如果你的代码是：
被 Nuitka、PyInstaller 等工具打包成了二进制（.exe/.pyd/.so），
或者是交互式环境、Jupyter、某些 Cython/Numba 编译的函数，
或者是 zipimport、egg、wheel 等非源码分发方式，
那么 inspect 就无法找到源码文件，自然会报 OSError: could not get source code。
PaddlePaddle 的 dy2static/jit 问题
PaddlePaddle 的动态图转静态（dy2static）和 JIT 导出模型时，会尝试用 inspect 获取模型的 forward 函数源码，做 AST 分析和转换。这在源码环境下没问题，但在打包环境下（如 Nuitka、PyInstaller）会100%失败。
但由于实际需要必须采用nuitka打包模式运行推理，应该如何修改可以让训练动态图模型导出为推理静态图模型，当前发现导出onnx模型也需要先转为静态图。

🏃‍♂️ Environment (运行环境)

paddlepaddle-gpu 2.4.2.post117（使用此版本原因是没有torch导入冲突且没有_dtype的paddle.tensor报错）
paddleocr 2.10.0
cuda 11.8
paddle2onnx 1.3.1

🌰 Minimal Reproducible Example (最小可复现问题的Demo)

python -m nuitka --standalone --include-package=paddleocr --include-package=utils --include-package=numpy --include-package=PIL --include-package=scipy --include-package=skimage --include-package=yaml --include-package=dicttoxml --include-package=xml --include-package=threading --include-package=queue --include-package=socket --include-package=logging --include-package=tempfile --include-package=platform --include-package=ctypes --include-package=glob --follow-imports --enable-plugin=numpy server.py

[Bobholamovic]

看起来涉及到飞桨框架动转静功能与打包工具的机制冲突，建议到飞桨框架仓库提问～