Existing project using v0.5.x fails due to functions in the model

未关闭
  • 哪里不对？
#2,933 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 4 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
45/100
Issue 类型
缺陷
描述清晰度
需要澄清
活跃度
冷清
技术栈
python, pytorch
领域
machine-learning

调研方向

使用 torch~=2.9.0 和 onnxscript~=0.5.6，仅通过 torch.nn.Linear 和 torch.onnx.export 重现该失败。检查依赖项规范和 _ConvertVersionPassRequiresInline 失败，然后验证已解析的依赖项允许在没有意外模型函数的情况下进行导出。

由索引模型根据 Issue 内容生成。

描述

New deployment of a project with dependencies torch~=2.9.0, onnxscript~=0.5.6 ran into errors on torch.onnx.export. The error is not correlated to the exporting module (exporting a sole torch.nn.Linear still causes error). Example output:

[torch.onnx] Translate the graph into ONNX...
[torch.onnx] Translate the graph into ONNX... ✅
Traceback (most recent call last):
  File "D:\Workspaces\Playgrounds\.venv\Lib\site-packages\onnx_ir\passes\_pass_infra.py", line 250, in call
    pass_result = pass_(model)
                  ^^^^^^^^^^^^
  File "D:\Workspaces\Playgrounds\.venv\Lib\site-packages\onnx_ir\passes\_pass_infra.py", line 127, in __call__
    result = self.call(model)
             ^^^^^^^^^^^^^^^^
  File "D:\Workspaces\Playgrounds\.venv\Lib\site-packages\onnxscript\version_converter\__init__.py", line 77, in call
    raise ValueError(
ValueError: The model contains functions. The version conversion pass does not support functions. Please use `common_passes.InlinePass` to inline the functions before applying this pass (_ConvertVersionPassRequiresInline).

The above exception was the direct cause of the following exception:
……
onnx_ir.passes.PassError: An error occurred when running the '<onnxscript.version_converter._ConvertVersionPassRequiresInline object at 0x00000194E90F6E70>' pass after the following passes: ['<onnx_ir.passes.common.inliner.InlinePass object at 0x00000194E2E0C770>']

Breaking into where the error is raised, inspecting pass_result.model.functions gives {('pkg.onnxscript.torch_lib.common', 'Rank', ''): Function('pkg.onnxscript.torch_lib.common', 'Rank', '', inputs=[Value(name='input')], attributes={}), outputs=[Value(name='return_val', producer='n1', index=0)]), ('pkg.onnxscript.torch_lib.common', 'IsScalar', ''): Function('pkg.onnxscript.torch_lib.common', 'IsScalar', '', inputs=[Value(name='input')], attributes={}), outputs=[Value(name='return_val', producer='n3', index=0)])}, which is not empty as expected.

It looks like a problem with the dependency specifications of onnxscript, as previous deployments with the same requirements.txt did not encounter such error.

Updating the package seems to clear the error but I guess it would be better if the dependencies could be fixed.

EDIT: Seems to be related to https://github.com/microsoft/onnxscript/issues/2790

主要语言
Python
星标
457
派生
149
平均合并
3 天 16 小时
30 天内合并 PR
24

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

microsoft/onnxscript 的其他 Issue

查看 microsoft/onnxscript 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。